Mango is a popular fruit for local consumption and export commodity. Currently, Indonesian mango export at 37.8 M accounted for 0.115% of world consumption. Pest and disease are the common enemies of mango that degrade the quality of mango yield. Specialized treatment in export destinations such as gamma-ray in Australia, or hot water treatment in Korea, demands pest-free and high-quality products. Artificial intelligence helps to improve mango pest and disease control. This paper compares the deep learning model on mango fruit pests and disease recognition. This research compares Visual Geometry Group 16 (VGG16), residual neural network 50 (ResNet50), InceptionResNet-V2, Inception-V3, and DenseNet architectures to identify pests and diseases on mango fruit. We implement transfer learning, adopt all pre-trained weight parameters from all those architectures, and replace the final layer to adjust the output. All the architectures are re-train and validated using our dataset. The tropical mango dataset is collected and labeled by a subject matter expert. The VGG16 model achieves the top validation and testing accuracy at 89% and 90%, respectively. VGG16 is the shallowest model, with 16 layers; therefore, the model was the smallest size. The testing time is superior to the rest of the experiment at 2 seconds for 130 testing images.
Early detection of tomato leaf diseases based on deep learning techniquesIAESIJAI
Tomato leaf diseases are a big issue for producers, and finding a single method to combat them is tough. Deep learning techniques, notably convolutional neural networks (CNNs), show promise in recognizing early indicators of illness, which can help producers avoid costly concerns in the future. In this study, we present a CNN-based model for the early identification of tomato leaf diseases to preserve output and boost yield. We used a dataset from the plantvillage database with 11,000 photos from 10 distinct disease categories to train our model. Our CNN was trained on this dataset, and the suggested model obtained an astounding 96% accuracy rate. This shows that our method has the potential to be efficient in detecting tomato leaf diseases early on, therefore assisting producers in managing and reducing disease outbreaks and, as a result, resulting in higher crop yields.
Cucumber disease recognition using machine learning and transfer learningriyaniaes
Cucumber is grown, as a cash crop besides it is one of the main and popular vegetables in Bangladesh. As Bangladesh's economy is largely dependent on the agricultural sector, cucumber farming could make economic and productivity growth more sustainable. But many diseases diminish the situation of cucumber. Early detection of disease can help to stop disease from spreading to other healthy plants and also accurate identifying the disease will help to reduce crop losses through specific treatments. In this paper, we have presented two approaches namely traditional machine learning (ML) and CNN-based transfer learning. Then we have compared the performance of the applied techniques to find out the most appropriate techniques for recognizing cucumber diseases. In our ML approach, the system involves five steps. After collecting the image, pre-processing is done by resizing, filtering, and contrast-enhancing. Then we have compared various ML algorithms using k-means based image segmentation after extracted 10 relevant features. Random forest gives the best accuracy with 89.93% in the traditional ML approach. We also studied and applied CNN-based transfer learning to investigate the further improvement of recognition performance. Lastly, a comparison among various transfer learning models such as InceptionV3, MobileNetV2, and VGG16 has been performed. Between these two approaches, MobileNetV2 achieves the highest accuracy with 93.23%.
Development of durian leaf disease detection on Android device IJECEIAES
Durian is exceedingly abundant in the Philippines, providing incomes for smallholder farmers. But amidst these things, durian is still vulnerable to different plant diseases that can cause significant economic loss in the agricultural industry. The conventional way of dealing plant disease detection is through naked-eye observation done by experts. To control such diseases using the old method is extensively laborious, time-consuming and costly especially in dealing with large fields. Hence, the proponent’s objective of this study is to create a standalone mobile app for durian leaf disease detection using the transfer learning approach. In this approach, the chosen network MobileNets, is pre-trained with a large scale of general datasets namely ImageNet to effective function as a generic template for visual processing. The pre-trained network transfers all the learned parameters and set as a feature extractor for the target task to be executed. Four health conditions are addressed in this study, 10 per classification with a total of 40 samples tested to evaluate the accuracy of the system. The result showed 90% in overall accuracy for detecting algalspot, cercospora, leaf discoloration and healthy leaf.
A deep learning-based approach for early detection of disease in sugarcane pl...IAESIJAI
In many regions of the nation, agriculture serves as the primary industry. The farming environment now faces a number of challenges to farmers. One of the major concerns, and the focus of this research, is disease prediction. A methodology is suggested to automate a process for identifying disease in plant growth and warning farmers in advance so they can take appropriate action. Disease in crop plants has an impact on agricultural production. In this work, a novel DenseNet-support vector machine: explainable artificial intelligence (DNet-SVM: XAI) interpretation that combines a DenseNet with support vector machine (SVM) and local interpretable model-agnostic explanation (LIME) interpretation has been proposed. DNet-SVM: XAI was created by a series of modifications to DenseNet201, including the addition of a support vector machine (SVM) classifier. Prior to using SVM to identify if an image is healthy or un-healthy, images are first feature extracted using a convolution network called DenseNet. In addition to offering a likely explanation for the prediction, the reasoning is carried out utilizing the visual cue produced by the LIME. In light of this, the proposed approach, when paired with its determined interpretability and precision, may successfully assist farmers in the detection of infected plants and recommendation of pesticide for the identified disease.
Low-cost convolutional neural network for tomato plant diseases classificationIAESIJAI
Agriculture is a crucial element to build a strong economy, not only because
of its importance in providing food, but also as a source of raw materials for
industry as well as source of energy. Different diseases affect plants, which
leads to decrease in productivity. In recent years, developments in
computing technology and machine-learning algorithms (such as deep neural
networks) in the field of agriculture have played a great role to face this
problem by building early detection tools. In this paper, we propose an
automatic plant disease classification based on a low complexity
convolutional neural network (CNN) architecture, which leads to faster online classification. For the training process, we used more than one 57,000
tomato leaf images representing nine classes, taken under natural
environment, and considered during training without background
subtraction. The designed model achieves 97.04% classification accuracy
and less than 0.2 error, which shows a high accuracy in distinguishing a
disease from another.
A deep learning-based mobile app system for visual identification of tomato p...IJECEIAES
Tomato is one of many horticulture crops in Indonesia which plays a vital role in supplying public food needs. However, tomato is a very susceptible plant to pests and diseases caused by bacteria and fungus. The infected diseases should be isolated as soon as it was detected. Therefore, developing a reliable and fast system is essential for controlling tomato pests and diseases. The deep learning-based application can help to speed up the identification of tomato disease as it can perform direct identification from the image. In this research, EfficientNetB0 was implemented to perform multi-class tomato plant disease classification. The model was then deployed to an android-based application using machine learning (ML) kit library. The proposed system obtained satisfactory results, reaching an average accuracy of 91.4%.
Plant Leaf Diseases Identification in Deep LearningCSEIJJournal
Crop diseases constitute a big threat to plant existence, but their rapid identification remains difficult in
many parts of the planet because of the shortage of the required infrastructure. In computer vision, plant
leaf detection made possible by deep learning has paved the way for smartphone-assisted disease
diagnosis. employing a public dataset of 4,306 images of diseased and healthy plant leaves collected under
controlled conditions, we train a deep convolutional neural network to spot one crop species and
4 diseases (or absence thereof). The trained model achieves an accuracy of 97.35% on a held-out test set,
demonstrating the feasibility of this approach. Overall, the approach of coaching deep learning models on
increasingly large and publicly available image datasets presents a transparent path toward smartphone-
assisted crop disease diagnosis on a large global scale. After the disease is successfully predicted with a
decent confidence level, the corresponding remedy for the disease present is displayed that may be taken as
a cure.
Early detection of tomato leaf diseases based on deep learning techniquesIAESIJAI
Tomato leaf diseases are a big issue for producers, and finding a single method to combat them is tough. Deep learning techniques, notably convolutional neural networks (CNNs), show promise in recognizing early indicators of illness, which can help producers avoid costly concerns in the future. In this study, we present a CNN-based model for the early identification of tomato leaf diseases to preserve output and boost yield. We used a dataset from the plantvillage database with 11,000 photos from 10 distinct disease categories to train our model. Our CNN was trained on this dataset, and the suggested model obtained an astounding 96% accuracy rate. This shows that our method has the potential to be efficient in detecting tomato leaf diseases early on, therefore assisting producers in managing and reducing disease outbreaks and, as a result, resulting in higher crop yields.
Cucumber disease recognition using machine learning and transfer learningriyaniaes
Cucumber is grown, as a cash crop besides it is one of the main and popular vegetables in Bangladesh. As Bangladesh's economy is largely dependent on the agricultural sector, cucumber farming could make economic and productivity growth more sustainable. But many diseases diminish the situation of cucumber. Early detection of disease can help to stop disease from spreading to other healthy plants and also accurate identifying the disease will help to reduce crop losses through specific treatments. In this paper, we have presented two approaches namely traditional machine learning (ML) and CNN-based transfer learning. Then we have compared the performance of the applied techniques to find out the most appropriate techniques for recognizing cucumber diseases. In our ML approach, the system involves five steps. After collecting the image, pre-processing is done by resizing, filtering, and contrast-enhancing. Then we have compared various ML algorithms using k-means based image segmentation after extracted 10 relevant features. Random forest gives the best accuracy with 89.93% in the traditional ML approach. We also studied and applied CNN-based transfer learning to investigate the further improvement of recognition performance. Lastly, a comparison among various transfer learning models such as InceptionV3, MobileNetV2, and VGG16 has been performed. Between these two approaches, MobileNetV2 achieves the highest accuracy with 93.23%.
Development of durian leaf disease detection on Android device IJECEIAES
Durian is exceedingly abundant in the Philippines, providing incomes for smallholder farmers. But amidst these things, durian is still vulnerable to different plant diseases that can cause significant economic loss in the agricultural industry. The conventional way of dealing plant disease detection is through naked-eye observation done by experts. To control such diseases using the old method is extensively laborious, time-consuming and costly especially in dealing with large fields. Hence, the proponent’s objective of this study is to create a standalone mobile app for durian leaf disease detection using the transfer learning approach. In this approach, the chosen network MobileNets, is pre-trained with a large scale of general datasets namely ImageNet to effective function as a generic template for visual processing. The pre-trained network transfers all the learned parameters and set as a feature extractor for the target task to be executed. Four health conditions are addressed in this study, 10 per classification with a total of 40 samples tested to evaluate the accuracy of the system. The result showed 90% in overall accuracy for detecting algalspot, cercospora, leaf discoloration and healthy leaf.
A deep learning-based approach for early detection of disease in sugarcane pl...IAESIJAI
In many regions of the nation, agriculture serves as the primary industry. The farming environment now faces a number of challenges to farmers. One of the major concerns, and the focus of this research, is disease prediction. A methodology is suggested to automate a process for identifying disease in plant growth and warning farmers in advance so they can take appropriate action. Disease in crop plants has an impact on agricultural production. In this work, a novel DenseNet-support vector machine: explainable artificial intelligence (DNet-SVM: XAI) interpretation that combines a DenseNet with support vector machine (SVM) and local interpretable model-agnostic explanation (LIME) interpretation has been proposed. DNet-SVM: XAI was created by a series of modifications to DenseNet201, including the addition of a support vector machine (SVM) classifier. Prior to using SVM to identify if an image is healthy or un-healthy, images are first feature extracted using a convolution network called DenseNet. In addition to offering a likely explanation for the prediction, the reasoning is carried out utilizing the visual cue produced by the LIME. In light of this, the proposed approach, when paired with its determined interpretability and precision, may successfully assist farmers in the detection of infected plants and recommendation of pesticide for the identified disease.
Low-cost convolutional neural network for tomato plant diseases classificationIAESIJAI
Agriculture is a crucial element to build a strong economy, not only because
of its importance in providing food, but also as a source of raw materials for
industry as well as source of energy. Different diseases affect plants, which
leads to decrease in productivity. In recent years, developments in
computing technology and machine-learning algorithms (such as deep neural
networks) in the field of agriculture have played a great role to face this
problem by building early detection tools. In this paper, we propose an
automatic plant disease classification based on a low complexity
convolutional neural network (CNN) architecture, which leads to faster online classification. For the training process, we used more than one 57,000
tomato leaf images representing nine classes, taken under natural
environment, and considered during training without background
subtraction. The designed model achieves 97.04% classification accuracy
and less than 0.2 error, which shows a high accuracy in distinguishing a
disease from another.
A deep learning-based mobile app system for visual identification of tomato p...IJECEIAES
Tomato is one of many horticulture crops in Indonesia which plays a vital role in supplying public food needs. However, tomato is a very susceptible plant to pests and diseases caused by bacteria and fungus. The infected diseases should be isolated as soon as it was detected. Therefore, developing a reliable and fast system is essential for controlling tomato pests and diseases. The deep learning-based application can help to speed up the identification of tomato disease as it can perform direct identification from the image. In this research, EfficientNetB0 was implemented to perform multi-class tomato plant disease classification. The model was then deployed to an android-based application using machine learning (ML) kit library. The proposed system obtained satisfactory results, reaching an average accuracy of 91.4%.
Plant Leaf Diseases Identification in Deep LearningCSEIJJournal
Crop diseases constitute a big threat to plant existence, but their rapid identification remains difficult in
many parts of the planet because of the shortage of the required infrastructure. In computer vision, plant
leaf detection made possible by deep learning has paved the way for smartphone-assisted disease
diagnosis. employing a public dataset of 4,306 images of diseased and healthy plant leaves collected under
controlled conditions, we train a deep convolutional neural network to spot one crop species and
4 diseases (or absence thereof). The trained model achieves an accuracy of 97.35% on a held-out test set,
demonstrating the feasibility of this approach. Overall, the approach of coaching deep learning models on
increasingly large and publicly available image datasets presents a transparent path toward smartphone-
assisted crop disease diagnosis on a large global scale. After the disease is successfully predicted with a
decent confidence level, the corresponding remedy for the disease present is displayed that may be taken as
a cure.
Peanut leaf spot disease identification using pre-trained deep convolutional...IJECEIAES
Reduction of quality and quantity of agricultural products, particularly peanut or groundnut, is usually associated with disease. This could be solved through automatic identification and diagnoses using deep learning. However, this technology is not yet explored and examined in the case of peanut leaf spot disease due to some aspects, such as the availability of sufficient data to be used for training and testing the model. This study is intended to explore the use of pre-trained visual geometry group–16 (VGG16), visual geometry group–19 (VGG19), InceptionV3, MobileNet, DenseNet, Xception, InceptionResNetV2, and ResNet50 architectures and deep learning optimizers such as stochastic gradient descent (SGD) with Momentum, adaptive moment estimation (Adam), root mean square propagation (RMSProp), and adaptive gradient algorithm (Adagrad) in creating a model that can identify leaf spot disease by using a total of 1,000 images of leaves captured using a mobile camera. Confusion matrix was used to assess the accuracy and precision of the results. The result of the study shows that DenseNet-169 trained using SGD with momentum, Adam, and RMSProp attained the highest accuracy of 98%, while DenseNet-169 trained using RMSProp achieved the highest precision of 98% among pre-trained deep convolutional neural network architectures. Furthermore, this result could be beneficial in agricultural automation and disease identification systems for peanut or groundnut plants.
Diseases in edible and industrial plants remains a major concern, affecting producers and consumers. The problem is further exacerbated as there are different species of plants with a wide variety of diseases that reduce the effectiveness of certain pesticides while increasing our risk of illness. A timely, accurate and automated detection of diseases can be beneficial. Our work focuses on evaluating deep learning (DL) approaches using transfer learning to automatically detect diseases in plants. To enhance the capabilities of our approach, we compiled a novel image dataset containing 87,570 records encompassing 32 different plants and 74 types of diseases. The dataset consists of leaf images from both laboratory setups and cultivation fields, making it more representative. To the best of our knowledge, no such datasets have been used for DL models. Four pretrained computer vision models, namely VGG-16, VGG-19, ResNet-50, and ResNet-101 were evaluated on our dataset. Our experiments demonstrate that both VGG-16 and VGG-19 models proved more efficient, yielding an accuracy of approximately 86% and a f1-score of 87%, as compared to ResNet-50 and ResNet-101. ResNet-50 attains an accuracy and a f1-score of 46.9% and 45.6%, respectively, while ResNet-101 reaches an accuracy of 40.7% and a f1-score of 26.9%.
Detection of diseases in rice leaf using convolutional neural network with tr...IJECEIAES
Rice is the most cultivated crop in the agricultural field and is a major food source for more than 60% of the population in India. The occurrence of disease in rice leaves, majorly affects the rice quality, production and also the farmers’ income. Nowadays, new variety of diseases in rice leaves are identified and detected periodically throughout the world. Manual monitoring and detection of plant diseases proves to be time consuming for the farmer and also a costly affair for using chemicals in the disease treatment. In this paper, a deep learning method of convolutional neural networks (CNN) with a transfer learning technique is proposed for the detection of a variety of diseases in rice leaves. This method uses the ResNeXt network model for classifying the images of disease-affected plants. The proposed model’s performance is evaluated using accuracy, precision, recall, F1-score and specificity. The experimental results of ResNeXt model measured for accuracy, precision, recall, F1-score, and specificity, are respectively 99.22%, 92.87%, 91.97%, 90.95%, and 99.05%, which proves greater accuracy improvement than the existing methods of SG2-ADA, YOLOV5, InceptionResNetV2 and Raspberry Pi.
Improved vision-based diagnosis of multi-plant disease using an ensemble of d...IJECEIAES
Farming and plants are crucial parts of the inward economy of a nation, which significantly boosts the economic growth of a country. Preserving plants from several disease infections at their early stage becomes cumbersome due to the absence of efficient diagnosis tools. Diverse difficulties lie in existing methods of plant disease recognition. As a result, developing a rapid and efficient multi-plant disease diagnosis system is a challenging task. At present, deep learning-based methods are frequently utilized for diagnosing plant diseases, which outperformed existing methods with higher efficiency. In order to investigate plant diseases more accurately, this article addresses an efficient hybrid approach using deep learning-based methods. Xception and ResNet50 models were applied for the classification of plant diseases, and these models were merged using the stacking ensemble learning technique to generate a hybrid model. A multi-plant dataset was created using leaf images of four plants: black gram, betel, Malabar spinach, and litchi, which contains nine classes and 44,972 images. Compared to existing individual convolutional neural networks (CNN) models, the proposed hybrid model is more feasible and effective, which acquired 99.20% accuracy. The outcomes and comparison with existing methods represent that the designed method can acquire competitive performance on the multi-plant disease diagnosis tasks.
Convolutional neural networks (CNN) trained using deep learning (DL) have advanced dramatically in recent years. Researchers from a variety of fields have been motivated by the success of CNNs in computer vision to develop better CNN models for use in other visually-rich settings. Successes in image classification and research have been achieved in a wide variety of domains throughout the past year. Among the many popularized image classification techniques, the detection of plant leaf diseases has received extensive research. As a result of the nature of the procedure, image quality is often degraded and distortions are introduced during the capturing of the image. In this study, we look into how various CNN models are affected by distortions. Corn-maze leaf photos from the 4,188-image corn or maize leaf Dataset (split into four categories) are under consideration. To evaluate how well they handle noise and blur, researchers have deployed pre-trained deep CNN models like visual geometry group (VGG), InceptionV3, ResNet50, and EfficientNetB0. Classification accuracy and metrics like as recall and f1-score are used to evaluate CNN performance.
Determination of Various Diseases in Two Most Consumed Fruits using Artificia...ijtsrd
Fruit diseases are manifested by deformations during or after harvesting the components in the fruit, when the infestation is caused by spores, fungi, insects or other contaminants. In early agricultural practices, it is thought that non destructive examination is possible with the analysis of pre harvest fruit leaves and early diagnosis of the disease, while post harvest detection and classification of fruit disease is possible by evaluating simple image processing techniques. Diseases of rotten or stained fruits without destruction. In this way, the disease will be identified and classified and the awareness of the producer for the next harvest will be provided. For this purpose, studies were carried out with apple and quince fruit, images were determined using still fruit pictures and machine learning, and disease classification was provided with labels. Image processing techniques are a system that detects disease made to a real time camera and prints it on the screen. Within the scope of this study, the data set was created and images of 22 apples and 18 quinces were taken. The image was classified by similarities in the literature review. The success of the proposed Convolutional Neural Network architecture in recognizing the disease was evaluated. By comparing the trained network, AlexNet architecture, with the proposed architecture, it has been determined that the success of image recognition has increased with the proposed method. Aysun Yilmaz Kizilboga | Atilla Ergüzen | Erdal Erdal "Determination of Various Diseases in Two Most Consumed Fruits using Artificial Neural Networks and Deep Learning Techniques" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-1 , December 2020, URL: https://www.ijtsrd.com/papers/ijtsrd38128.pdf Paper URL : https://www.ijtsrd.com/engineering/computer-engineering/38128/determination-of-various-diseases-in-two-most-consumed-fruits-using-artificial-neural-networks-and-deep-learning-techniques/aysun-yilmaz-kizilboga
Using deep learning algorithms to classify crop diseasesIJECEIAES
The use of deep learning algorithms for the classification of crop diseases is one of the promising areas in agricultural technology. This is due to the need for rapid and accurate detection of plant diseases, which allows timely measures to be taken to treat them and prevent their spread. One of them is to increase productivity and maintain land quality through the timely detection of diseases and pests in agriculture and their elimination. Traditional classification methods in machine learning and algorithms in deep learning were compared to note the high accuracy in detecting pests and crop diseases. The advantages and disadvantages of each model considered during training were taken into account, and the Inception V3 algorithm was incorporated into the application. They can monitor the condition of crops on a daily basis with the help of new technology-applications on gadgets. Aerial photographs used by research institutes and agricultural grain centers do not show the changes that occur in agricultural grains, that is, diseases and pests. Therefore, the method proposed in this paper determines the types of diseases and pests of cereals through a mobile application and suggests ways to deal with them.
Robusta coffee leaf diseases detection based on MobileNetV2 modelIJECEIAES
Indonesia is a major exporter and producer of coffee, and coffee cultivation adds to the nation's economy. Despite this, coffee remains vulnerable to several plant diseases that may result in significant financial losses for the agricultural industry. Traditionally, plant diseases are detected by expert observation with the naked eye. Traditional methods for managing such diseases are arduous, time-consuming, and costly, especially when dealing with expansive territories. Using a model based on transfer learning and deep learning model, we present in this study a technique for classifying Robusta coffee leaf disease photos into healthy and unhealthy classes. The MobileNetV2 network serves as the model since its network design is simple. Therefore, it is likely that the suggested approach will be deployed further on mobile devices. In addition, the transfer learning and experimental learning paradigms. Because it is such a lightweight net, the MobileNetV2 system serves as the foundational model. Results on Robusta coffee leaf disease datasets indicate that the suggested technique can achieve a high level of accuracy, up to 99.93%. The accuracy of other architectures besides MobileNetV2 such as DenseNet169 is 99.74%, ResNet50 architecture is 99.41%, and InceptionResNetV2 architecture is 99.09%.
Corn plant disease classification based on leaf using residual networks-9 arc...IJECEIAES
Corn plants are classified based on the leaf as healthy leafy and have 3 types of diseases leaf namely northern leaf blight, common rust, and gray leaf spot. Convolutional neural network (CNN) is the most popular structure for classification image detection. In this study, ResNet-9 architecture was implemented to build the best model CNN for the classification of corn plant diseases. After that, we do comparisons of epochs 5, 25, 55, 75, and 100 to get the best model. The highest accuracy value was obtained in the 100epoch experiment so in this study 100 epochs were used in model formation. The dataset source in this study uses a dataset taken from the Kaggle platform as many as 9145 leaf corn plant data which is divided into training data (80%) and testing data (80%). In this study, three hyperparameter tuning experiments were carried out and the results of hyperparameter tuning experiments where num_workers is 4 and batch_size is 32. This classification obtained an accuracy rate of 99% and the model is implemented into a web interface.
Amazon products reviews classification based on machine learning, deep learni...TELKOMNIKA JOURNAL
In recent times, the trend of online shopping through e-commerce stores and websites has grown to a huge extent. Whenever a product is purchased on an e-commerce platform, people leave their reviews about the product. These reviews are very helpful for the store owners and the product’s manufacturers for the betterment of their work process as well as product quality. An automated system is proposed in this work that operates on two datasets D1 and D2 obtained from Amazon. After certain preprocessing steps, N-gram and word embedding-based features are extracted using term frequency-inverse document frequency (TF-IDF), bag of words (BoW) and global vectors (GloVe), and Word2vec, respectively. Four machine learning (ML) models support vector machines (SVM), logistic regression (RF), logistic regression (LR), multinomial Naïve Bayes (MNB), two deep learning (DL) models convolutional neural network (CNN), long-short term memory (LSTM), and standalone bidirectional encoder representations (BERT) are used to classify reviews as either positive or negative. The results obtained by the standard ML, DL models and BERT are evaluated using certain performance evaluation measures. BERT turns out to be the best-performing model in the case of D1 with an accuracy of 90% on features derived by word embedding models while the CNN provides the best accuracy of 97% upon word embedding features in the case of D2. The proposed model shows better overall performance on D2 as compared to D1.
Design, simulation, and analysis of microstrip patch antenna for wireless app...TELKOMNIKA JOURNAL
In this study, a microstrip patch antenna that works at 3.6 GHz was built and tested to see how well it works. In this work, Rogers RT/Duroid 5880 has been used as the substrate material, with a dielectric permittivity of 2.2 and a thickness of 0.3451 mm; it serves as the base for the examined antenna. The computer simulation technology (CST) studio suite is utilized to show the recommended antenna design. The goal of this study was to get a more extensive transmission capacity, a lower voltage standing wave ratio (VSWR), and a lower return loss, but the main goal was to get a higher gain, directivity, and efficiency. After simulation, the return loss, gain, directivity, bandwidth, and efficiency of the supplied antenna are found to be -17.626 dB, 9.671 dBi, 9.924 dBi, 0.2 GHz, and 97.45%, respectively. Besides, the recreation uncovered that the transfer speed side-lobe level at phi was much better than those of the earlier works, at -28.8 dB, respectively. Thus, it makes a solid contender for remote innovation and more robust communication.
More Related Content
Similar to A comparative study of mango fruit pest and disease recognition
Peanut leaf spot disease identification using pre-trained deep convolutional...IJECEIAES
Reduction of quality and quantity of agricultural products, particularly peanut or groundnut, is usually associated with disease. This could be solved through automatic identification and diagnoses using deep learning. However, this technology is not yet explored and examined in the case of peanut leaf spot disease due to some aspects, such as the availability of sufficient data to be used for training and testing the model. This study is intended to explore the use of pre-trained visual geometry group–16 (VGG16), visual geometry group–19 (VGG19), InceptionV3, MobileNet, DenseNet, Xception, InceptionResNetV2, and ResNet50 architectures and deep learning optimizers such as stochastic gradient descent (SGD) with Momentum, adaptive moment estimation (Adam), root mean square propagation (RMSProp), and adaptive gradient algorithm (Adagrad) in creating a model that can identify leaf spot disease by using a total of 1,000 images of leaves captured using a mobile camera. Confusion matrix was used to assess the accuracy and precision of the results. The result of the study shows that DenseNet-169 trained using SGD with momentum, Adam, and RMSProp attained the highest accuracy of 98%, while DenseNet-169 trained using RMSProp achieved the highest precision of 98% among pre-trained deep convolutional neural network architectures. Furthermore, this result could be beneficial in agricultural automation and disease identification systems for peanut or groundnut plants.
Diseases in edible and industrial plants remains a major concern, affecting producers and consumers. The problem is further exacerbated as there are different species of plants with a wide variety of diseases that reduce the effectiveness of certain pesticides while increasing our risk of illness. A timely, accurate and automated detection of diseases can be beneficial. Our work focuses on evaluating deep learning (DL) approaches using transfer learning to automatically detect diseases in plants. To enhance the capabilities of our approach, we compiled a novel image dataset containing 87,570 records encompassing 32 different plants and 74 types of diseases. The dataset consists of leaf images from both laboratory setups and cultivation fields, making it more representative. To the best of our knowledge, no such datasets have been used for DL models. Four pretrained computer vision models, namely VGG-16, VGG-19, ResNet-50, and ResNet-101 were evaluated on our dataset. Our experiments demonstrate that both VGG-16 and VGG-19 models proved more efficient, yielding an accuracy of approximately 86% and a f1-score of 87%, as compared to ResNet-50 and ResNet-101. ResNet-50 attains an accuracy and a f1-score of 46.9% and 45.6%, respectively, while ResNet-101 reaches an accuracy of 40.7% and a f1-score of 26.9%.
Detection of diseases in rice leaf using convolutional neural network with tr...IJECEIAES
Rice is the most cultivated crop in the agricultural field and is a major food source for more than 60% of the population in India. The occurrence of disease in rice leaves, majorly affects the rice quality, production and also the farmers’ income. Nowadays, new variety of diseases in rice leaves are identified and detected periodically throughout the world. Manual monitoring and detection of plant diseases proves to be time consuming for the farmer and also a costly affair for using chemicals in the disease treatment. In this paper, a deep learning method of convolutional neural networks (CNN) with a transfer learning technique is proposed for the detection of a variety of diseases in rice leaves. This method uses the ResNeXt network model for classifying the images of disease-affected plants. The proposed model’s performance is evaluated using accuracy, precision, recall, F1-score and specificity. The experimental results of ResNeXt model measured for accuracy, precision, recall, F1-score, and specificity, are respectively 99.22%, 92.87%, 91.97%, 90.95%, and 99.05%, which proves greater accuracy improvement than the existing methods of SG2-ADA, YOLOV5, InceptionResNetV2 and Raspberry Pi.
Improved vision-based diagnosis of multi-plant disease using an ensemble of d...IJECEIAES
Farming and plants are crucial parts of the inward economy of a nation, which significantly boosts the economic growth of a country. Preserving plants from several disease infections at their early stage becomes cumbersome due to the absence of efficient diagnosis tools. Diverse difficulties lie in existing methods of plant disease recognition. As a result, developing a rapid and efficient multi-plant disease diagnosis system is a challenging task. At present, deep learning-based methods are frequently utilized for diagnosing plant diseases, which outperformed existing methods with higher efficiency. In order to investigate plant diseases more accurately, this article addresses an efficient hybrid approach using deep learning-based methods. Xception and ResNet50 models were applied for the classification of plant diseases, and these models were merged using the stacking ensemble learning technique to generate a hybrid model. A multi-plant dataset was created using leaf images of four plants: black gram, betel, Malabar spinach, and litchi, which contains nine classes and 44,972 images. Compared to existing individual convolutional neural networks (CNN) models, the proposed hybrid model is more feasible and effective, which acquired 99.20% accuracy. The outcomes and comparison with existing methods represent that the designed method can acquire competitive performance on the multi-plant disease diagnosis tasks.
Convolutional neural networks (CNN) trained using deep learning (DL) have advanced dramatically in recent years. Researchers from a variety of fields have been motivated by the success of CNNs in computer vision to develop better CNN models for use in other visually-rich settings. Successes in image classification and research have been achieved in a wide variety of domains throughout the past year. Among the many popularized image classification techniques, the detection of plant leaf diseases has received extensive research. As a result of the nature of the procedure, image quality is often degraded and distortions are introduced during the capturing of the image. In this study, we look into how various CNN models are affected by distortions. Corn-maze leaf photos from the 4,188-image corn or maize leaf Dataset (split into four categories) are under consideration. To evaluate how well they handle noise and blur, researchers have deployed pre-trained deep CNN models like visual geometry group (VGG), InceptionV3, ResNet50, and EfficientNetB0. Classification accuracy and metrics like as recall and f1-score are used to evaluate CNN performance.
Determination of Various Diseases in Two Most Consumed Fruits using Artificia...ijtsrd
Fruit diseases are manifested by deformations during or after harvesting the components in the fruit, when the infestation is caused by spores, fungi, insects or other contaminants. In early agricultural practices, it is thought that non destructive examination is possible with the analysis of pre harvest fruit leaves and early diagnosis of the disease, while post harvest detection and classification of fruit disease is possible by evaluating simple image processing techniques. Diseases of rotten or stained fruits without destruction. In this way, the disease will be identified and classified and the awareness of the producer for the next harvest will be provided. For this purpose, studies were carried out with apple and quince fruit, images were determined using still fruit pictures and machine learning, and disease classification was provided with labels. Image processing techniques are a system that detects disease made to a real time camera and prints it on the screen. Within the scope of this study, the data set was created and images of 22 apples and 18 quinces were taken. The image was classified by similarities in the literature review. The success of the proposed Convolutional Neural Network architecture in recognizing the disease was evaluated. By comparing the trained network, AlexNet architecture, with the proposed architecture, it has been determined that the success of image recognition has increased with the proposed method. Aysun Yilmaz Kizilboga | Atilla Ergüzen | Erdal Erdal "Determination of Various Diseases in Two Most Consumed Fruits using Artificial Neural Networks and Deep Learning Techniques" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-1 , December 2020, URL: https://www.ijtsrd.com/papers/ijtsrd38128.pdf Paper URL : https://www.ijtsrd.com/engineering/computer-engineering/38128/determination-of-various-diseases-in-two-most-consumed-fruits-using-artificial-neural-networks-and-deep-learning-techniques/aysun-yilmaz-kizilboga
Using deep learning algorithms to classify crop diseasesIJECEIAES
The use of deep learning algorithms for the classification of crop diseases is one of the promising areas in agricultural technology. This is due to the need for rapid and accurate detection of plant diseases, which allows timely measures to be taken to treat them and prevent their spread. One of them is to increase productivity and maintain land quality through the timely detection of diseases and pests in agriculture and their elimination. Traditional classification methods in machine learning and algorithms in deep learning were compared to note the high accuracy in detecting pests and crop diseases. The advantages and disadvantages of each model considered during training were taken into account, and the Inception V3 algorithm was incorporated into the application. They can monitor the condition of crops on a daily basis with the help of new technology-applications on gadgets. Aerial photographs used by research institutes and agricultural grain centers do not show the changes that occur in agricultural grains, that is, diseases and pests. Therefore, the method proposed in this paper determines the types of diseases and pests of cereals through a mobile application and suggests ways to deal with them.
Robusta coffee leaf diseases detection based on MobileNetV2 modelIJECEIAES
Indonesia is a major exporter and producer of coffee, and coffee cultivation adds to the nation's economy. Despite this, coffee remains vulnerable to several plant diseases that may result in significant financial losses for the agricultural industry. Traditionally, plant diseases are detected by expert observation with the naked eye. Traditional methods for managing such diseases are arduous, time-consuming, and costly, especially when dealing with expansive territories. Using a model based on transfer learning and deep learning model, we present in this study a technique for classifying Robusta coffee leaf disease photos into healthy and unhealthy classes. The MobileNetV2 network serves as the model since its network design is simple. Therefore, it is likely that the suggested approach will be deployed further on mobile devices. In addition, the transfer learning and experimental learning paradigms. Because it is such a lightweight net, the MobileNetV2 system serves as the foundational model. Results on Robusta coffee leaf disease datasets indicate that the suggested technique can achieve a high level of accuracy, up to 99.93%. The accuracy of other architectures besides MobileNetV2 such as DenseNet169 is 99.74%, ResNet50 architecture is 99.41%, and InceptionResNetV2 architecture is 99.09%.
Corn plant disease classification based on leaf using residual networks-9 arc...IJECEIAES
Corn plants are classified based on the leaf as healthy leafy and have 3 types of diseases leaf namely northern leaf blight, common rust, and gray leaf spot. Convolutional neural network (CNN) is the most popular structure for classification image detection. In this study, ResNet-9 architecture was implemented to build the best model CNN for the classification of corn plant diseases. After that, we do comparisons of epochs 5, 25, 55, 75, and 100 to get the best model. The highest accuracy value was obtained in the 100epoch experiment so in this study 100 epochs were used in model formation. The dataset source in this study uses a dataset taken from the Kaggle platform as many as 9145 leaf corn plant data which is divided into training data (80%) and testing data (80%). In this study, three hyperparameter tuning experiments were carried out and the results of hyperparameter tuning experiments where num_workers is 4 and batch_size is 32. This classification obtained an accuracy rate of 99% and the model is implemented into a web interface.
Amazon products reviews classification based on machine learning, deep learni...TELKOMNIKA JOURNAL
In recent times, the trend of online shopping through e-commerce stores and websites has grown to a huge extent. Whenever a product is purchased on an e-commerce platform, people leave their reviews about the product. These reviews are very helpful for the store owners and the product’s manufacturers for the betterment of their work process as well as product quality. An automated system is proposed in this work that operates on two datasets D1 and D2 obtained from Amazon. After certain preprocessing steps, N-gram and word embedding-based features are extracted using term frequency-inverse document frequency (TF-IDF), bag of words (BoW) and global vectors (GloVe), and Word2vec, respectively. Four machine learning (ML) models support vector machines (SVM), logistic regression (RF), logistic regression (LR), multinomial Naïve Bayes (MNB), two deep learning (DL) models convolutional neural network (CNN), long-short term memory (LSTM), and standalone bidirectional encoder representations (BERT) are used to classify reviews as either positive or negative. The results obtained by the standard ML, DL models and BERT are evaluated using certain performance evaluation measures. BERT turns out to be the best-performing model in the case of D1 with an accuracy of 90% on features derived by word embedding models while the CNN provides the best accuracy of 97% upon word embedding features in the case of D2. The proposed model shows better overall performance on D2 as compared to D1.
Design, simulation, and analysis of microstrip patch antenna for wireless app...TELKOMNIKA JOURNAL
In this study, a microstrip patch antenna that works at 3.6 GHz was built and tested to see how well it works. In this work, Rogers RT/Duroid 5880 has been used as the substrate material, with a dielectric permittivity of 2.2 and a thickness of 0.3451 mm; it serves as the base for the examined antenna. The computer simulation technology (CST) studio suite is utilized to show the recommended antenna design. The goal of this study was to get a more extensive transmission capacity, a lower voltage standing wave ratio (VSWR), and a lower return loss, but the main goal was to get a higher gain, directivity, and efficiency. After simulation, the return loss, gain, directivity, bandwidth, and efficiency of the supplied antenna are found to be -17.626 dB, 9.671 dBi, 9.924 dBi, 0.2 GHz, and 97.45%, respectively. Besides, the recreation uncovered that the transfer speed side-lobe level at phi was much better than those of the earlier works, at -28.8 dB, respectively. Thus, it makes a solid contender for remote innovation and more robust communication.
Design and simulation an optimal enhanced PI controller for congestion avoida...TELKOMNIKA JOURNAL
In this paper, snake optimization algorithm (SOA) is used to find the optimal gains of an enhanced controller for controlling congestion problem in computer networks. M-file and Simulink platform is adopted to evaluate the response of the active queue management (AQM) system, a comparison with two classical controllers is done, all tuned gains of controllers are obtained using SOA method and the fitness function chose to monitor the system performance is the integral time absolute error (ITAE). Transient analysis and robust analysis is used to show the proposed controller performance, two robustness tests are applied to the AQM system, one is done by varying the size of queue value in different period and the other test is done by changing the number of transmission control protocol (TCP) sessions with a value of ± 20% from its original value. The simulation results reflect a stable and robust behavior and best performance is appeared clearly to achieve the desired queue size without any noise or any transmission problems.
Improving the detection of intrusion in vehicular ad-hoc networks with modifi...TELKOMNIKA JOURNAL
Vehicular ad-hoc networks (VANETs) are wireless-equipped vehicles that form networks along the road. The security of this network has been a major challenge. The identity-based cryptosystem (IBC) previously used to secure the networks suffers from membership authentication security features. This paper focuses on improving the detection of intruders in VANETs with a modified identity-based cryptosystem (MIBC). The MIBC is developed using a non-singular elliptic curve with Lagrange interpolation. The public key of vehicles and roadside units on the network are derived from number plates and location identification numbers, respectively. Pseudo-identities are used to mask the real identity of users to preserve their privacy. The membership authentication mechanism ensures that only valid and authenticated members of the network are allowed to join the network. The performance of the MIBC is evaluated using intrusion detection ratio (IDR) and computation time (CT) and then validated with the existing IBC. The result obtained shows that the MIBC recorded an IDR of 99.3% against 94.3% obtained for the existing identity-based cryptosystem (EIBC) for 140 unregistered vehicles attempting to intrude on the network. The MIBC shows lower CT values of 1.17 ms against 1.70 ms for EIBC. The MIBC can be used to improve the security of VANETs.
Conceptual model of internet banking adoption with perceived risk and trust f...TELKOMNIKA JOURNAL
Understanding the primary factors of internet banking (IB) acceptance is critical for both banks and users; nevertheless, our knowledge of the role of users’ perceived risk and trust in IB adoption is limited. As a result, we develop a conceptual model by incorporating perceived risk and trust into the technology acceptance model (TAM) theory toward the IB. The proper research emphasized that the most essential component in explaining IB adoption behavior is behavioral intention to use IB adoption. TAM is helpful for figuring out how elements that affect IB adoption are connected to one another. According to previous literature on IB and the use of such technology in Iraq, one has to choose a theoretical foundation that may justify the acceptance of IB from the customer’s perspective. The conceptual model was therefore constructed using the TAM as a foundation. Furthermore, perceived risk and trust were added to the TAM dimensions as external factors. The key objective of this work was to extend the TAM to construct a conceptual model for IB adoption and to get sufficient theoretical support from the existing literature for the essential elements and their relationships in order to unearth new insights about factors responsible for IB adoption.
Efficient combined fuzzy logic and LMS algorithm for smart antennaTELKOMNIKA JOURNAL
The smart antennas are broadly used in wireless communication. The least mean square (LMS) algorithm is a procedure that is concerned in controlling the smart antenna pattern to accommodate specified requirements such as steering the beam toward the desired signal, in addition to placing the deep nulls in the direction of unwanted signals. The conventional LMS (C-LMS) has some drawbacks like slow convergence speed besides high steady state fluctuation error. To overcome these shortcomings, the present paper adopts an adaptive fuzzy control step size least mean square (FC-LMS) algorithm to adjust its step size. Computer simulation outcomes illustrate that the given model has fast convergence rate as well as low mean square error steady state.
Design and implementation of a LoRa-based system for warning of forest fireTELKOMNIKA JOURNAL
This paper presents the design and implementation of a forest fire monitoring and warning system based on long range (LoRa) technology, a novel ultra-low power consumption and long-range wireless communication technology for remote sensing applications. The proposed system includes a wireless sensor network that records environmental parameters such as temperature, humidity, wind speed, and carbon dioxide (CO2) concentration in the air, as well as taking infrared photos.The data collected at each sensor node will be transmitted to the gateway via LoRa wireless transmission. Data will be collected, processed, and uploaded to a cloud database at the gateway. An Android smartphone application that allows anyone to easily view the recorded data has been developed. When a fire is detected, the system will sound a siren and send a warning message to the responsible personnel, instructing them to take appropriate action. Experiments in Tram Chim Park, Vietnam, have been conducted to verify and evaluate the operation of the system.
Wavelet-based sensing technique in cognitive radio networkTELKOMNIKA JOURNAL
Cognitive radio is a smart radio that can change its transmitter parameter based on interaction with the environment in which it operates. The demand for frequency spectrum is growing due to a big data issue as many Internet of Things (IoT) devices are in the network. Based on previous research, most frequency spectrum was used, but some spectrums were not used, called spectrum hole. Energy detection is one of the spectrum sensing methods that has been frequently used since it is easy to use and does not require license users to have any prior signal understanding. But this technique is incapable of detecting at low signal-to-noise ratio (SNR) levels. Therefore, the wavelet-based sensing is proposed to overcome this issue and detect spectrum holes. The main objective of this work is to evaluate the performance of wavelet-based sensing and compare it with the energy detection technique. The findings show that the percentage of detection in wavelet-based sensing is 83% higher than energy detection performance. This result indicates that the wavelet-based sensing has higher precision in detection and the interference towards primary user can be decreased.
A novel compact dual-band bandstop filter with enhanced rejection bandsTELKOMNIKA JOURNAL
In this paper, we present the design of a new wide dual-band bandstop filter (DBBSF) using nonuniform transmission lines. The method used to design this filter is to replace conventional uniform transmission lines with nonuniform lines governed by a truncated Fourier series. Based on how impedances are profiled in the proposed DBBSF structure, the fractional bandwidths of the two 10 dB-down rejection bands are widened to 39.72% and 52.63%, respectively, and the physical size has been reduced compared to that of the filter with the uniform transmission lines. The results of the electromagnetic (EM) simulation support the obtained analytical response and show an improved frequency behavior.
Deep learning approach to DDoS attack with imbalanced data at the application...TELKOMNIKA JOURNAL
A distributed denial of service (DDoS) attack is where one or more computers attack or target a server computer, by flooding internet traffic to the server. As a result, the server cannot be accessed by legitimate users. A result of this attack causes enormous losses for a company because it can reduce the level of user trust, and reduce the company’s reputation to lose customers due to downtime. One of the services at the application layer that can be accessed by users is a web-based lightweight directory access protocol (LDAP) service that can provide safe and easy services to access directory applications. We used a deep learning approach to detect DDoS attacks on the CICDDoS 2019 dataset on a complex computer network at the application layer to get fast and accurate results for dealing with unbalanced data. Based on the results obtained, it is observed that DDoS attack detection using a deep learning approach on imbalanced data performs better when implemented using synthetic minority oversampling technique (SMOTE) method for binary classes. On the other hand, the proposed deep learning approach performs better for detecting DDoS attacks in multiclass when implemented using the adaptive synthetic (ADASYN) method.
The appearance of uncertainties and disturbances often effects the characteristics of either linear or nonlinear systems. Plus, the stabilization process may be deteriorated thus incurring a catastrophic effect to the system performance. As such, this manuscript addresses the concept of matching condition for the systems that are suffering from miss-match uncertainties and exogeneous disturbances. The perturbation towards the system at hand is assumed to be known and unbounded. To reach this outcome, uncertainties and their classifications are reviewed thoroughly. The structural matching condition is proposed and tabulated in the proposition 1. Two types of mathematical expressions are presented to distinguish the system with matched uncertainty and the system with miss-matched uncertainty. Lastly, two-dimensional numerical expressions are provided to practice the proposed proposition. The outcome shows that matching condition has the ability to change the system to a design-friendly model for asymptotic stabilization.
Implementation of FinFET technology based low power 4×4 Wallace tree multipli...TELKOMNIKA JOURNAL
Many systems, including digital signal processors, finite impulse response (FIR) filters, application-specific integrated circuits, and microprocessors, use multipliers. The demand for low power multipliers is gradually rising day by day in the current technological trend. In this study, we describe a 4×4 Wallace multiplier based on a carry select adder (CSA) that uses less power and has a better power delay product than existing multipliers. HSPICE tool at 16 nm technology is used to simulate the results. In comparison to the traditional CSA-based multiplier, which has a power consumption of 1.7 µW and power delay product (PDP) of 57.3 fJ, the results demonstrate that the Wallace multiplier design employing CSA with first zero finding logic (FZF) logic has the lowest power consumption of 1.4 µW and PDP of 27.5 fJ.
Evaluation of the weighted-overlap add model with massive MIMO in a 5G systemTELKOMNIKA JOURNAL
The flaw in 5G orthogonal frequency division multiplexing (OFDM) becomes apparent in high-speed situations. Because the doppler effect causes frequency shifts, the orthogonality of OFDM subcarriers is broken, lowering both their bit error rate (BER) and throughput output. As part of this research, we use a novel design that combines massive multiple input multiple output (MIMO) and weighted overlap and add (WOLA) to improve the performance of 5G systems. To determine which design is superior, throughput and BER are calculated for both the proposed design and OFDM. The results of the improved system show a massive improvement in performance ver the conventional system and significant improvements with massive MIMO, including the best throughput and BER. When compared to conventional systems, the improved system has a throughput that is around 22% higher and the best performance in terms of BER, but it still has around 25% less error than OFDM.
Reflector antenna design in different frequencies using frequency selective s...TELKOMNIKA JOURNAL
In this study, it is aimed to obtain two different asymmetric radiation patterns obtained from antennas in the shape of the cross-section of a parabolic reflector (fan blade type antennas) and antennas with cosecant-square radiation characteristics at two different frequencies from a single antenna. For this purpose, firstly, a fan blade type antenna design will be made, and then the reflective surface of this antenna will be completed to the shape of the reflective surface of the antenna with the cosecant-square radiation characteristic with the frequency selective surface designed to provide the characteristics suitable for the purpose. The frequency selective surface designed and it provides the perfect transmission as possible at 4 GHz operating frequency, while it will act as a band-quenching filter for electromagnetic waves at 5 GHz operating frequency and will be a reflective surface. Thanks to this frequency selective surface to be used as a reflective surface in the antenna, a fan blade type radiation characteristic at 4 GHz operating frequency will be obtained, while a cosecant-square radiation characteristic at 5 GHz operating frequency will be obtained.
Reagentless iron detection in water based on unclad fiber optical sensorTELKOMNIKA JOURNAL
A simple and low-cost fiber based optical sensor for iron detection is demonstrated in this paper. The sensor head consist of an unclad optical fiber with the unclad length of 1 cm and it has a straight structure. Results obtained shows a linear relationship between the output light intensity and iron concentration, illustrating the functionality of this iron optical sensor. Based on the experimental results, the sensitivity and linearity are achieved at 0.0328/ppm and 0.9824 respectively at the wavelength of 690 nm. With the same wavelength, other performance parameters are also studied. Resolution and limit of detection (LOD) are found to be 0.3049 ppm and 0.0755 ppm correspondingly. This iron sensor is advantageous in that it does not require any reagent for detection, enabling it to be simpler and cost-effective in the implementation of the iron sensing.
Impact of CuS counter electrode calcination temperature on quantum dot sensit...TELKOMNIKA JOURNAL
In place of the commercial Pt electrode used in quantum sensitized solar cells, the low-cost CuS cathode is created using electrophoresis. High resolution scanning electron microscopy and X-ray diffraction were used to analyze the structure and morphology of structural cubic samples with diameters ranging from 40 nm to 200 nm. The conversion efficiency of solar cells is significantly impacted by the calcination temperatures of cathodes at 100 °C, 120 °C, 150 °C, and 180 °C under vacuum. The fluorine doped tin oxide (FTO)/CuS cathode electrode reached a maximum efficiency of 3.89% when it was calcined at 120 °C. Compared to other temperature combinations, CuS nanoparticles crystallize at 120 °C, which lowers resistance while increasing electron lifetime.
In place of the commercial Pt electrode used in quantum sensitized solar cells, the low-cost CuS cathode is created using electrophoresis. High resolution scanning electron microscopy and X-ray diffraction were used to analyze the structure and morphology of structural cubic samples with diameters ranging from 40 nm to 200 nm. The conversion efficiency of solar cells is significantly impacted by the calcination temperatures of cathodes at 100 °C, 120 °C, 150 °C, and 180 °C under vacuum. The fluorine doped tin oxide (FTO)/CuS cathode electrode reached a maximum efficiency of 3.89% when it was calcined at 120 °C. Compared to other temperature combinations, CuS nanoparticles crystallize at 120 °C, which lowers resistance while increasing electron lifetime.
A progressive learning for structural tolerance online sequential extreme lea...TELKOMNIKA JOURNAL
This article discusses the progressive learning for structural tolerance online sequential extreme learning machine (PSTOS-ELM). PSTOS-ELM can save robust accuracy while updating the new data and the new class data on the online training situation. The robustness accuracy arises from using the householder block exact QR decomposition recursive least squares (HBQRD-RLS) of the PSTOS-ELM. This method is suitable for applications that have data streaming and often have new class data. Our experiment compares the PSTOS-ELM accuracy and accuracy robustness while data is updating with the batch-extreme learning machine (ELM) and structural tolerance online sequential extreme learning machine (STOS-ELM) that both must retrain the data in a new class data case. The experimental results show that PSTOS-ELM has accuracy and robustness comparable to ELM and STOS-ELM while also can update new class data immediately.
Electroencephalography-based brain-computer interface using neural networksTELKOMNIKA JOURNAL
This study aimed to develop a brain-computer interface that can control an electric wheelchair using electroencephalography (EEG) signals. First, we used the Mind Wave Mobile 2 device to capture raw EEG signals from the surface of the scalp. The signals were transformed into the frequency domain using fast Fourier transform (FFT) and filtered to monitor changes in attention and relaxation. Next, we performed time and frequency domain analyses to identify features for five eye gestures: opened, closed, blink per second, double blink, and lookup. The base state was the opened-eyes gesture, and we compared the features of the remaining four action gestures to the base state to identify potential gestures. We then built a multilayer neural network to classify these features into five signals that control the wheelchair’s movement. Finally, we designed an experimental wheelchair system to test the effectiveness of the proposed approach. The results demonstrate that the EEG classification was highly accurate and computationally efficient. Moreover, the average performance of the brain-controlled wheelchair system was over 75% across different individuals, which suggests the feasibility of this approach.
Adaptive segmentation algorithm based on level set model in medical imagingTELKOMNIKA JOURNAL
For image segmentation, level set models are frequently employed. It offer best solution to overcome the main limitations of deformable parametric models. However, the challenge when applying those models in medical images stills deal with removing blurs in image edges which directly affects the edge indicator function, leads to not adaptively segmenting images and causes a wrong analysis of pathologies wich prevents to conclude a correct diagnosis. To overcome such issues, an effective process is suggested by simultaneously modelling and solving systems’ two-dimensional partial differential equations (PDE). The first PDE equation allows restoration using Euler’s equation similar to an anisotropic smoothing based on a regularized Perona and Malik filter that eliminates noise while preserving edge information in accordance with detected contours in the second equation that segments the image based on the first equation solutions. This approach allows developing a new algorithm which overcome the studied model drawbacks. Results of the proposed method give clear segments that can be applied to any application. Experiments on many medical images in particular blurry images with high information losses, demonstrate that the developed approach produces superior segmentation results in terms of quantity and quality compared to other models already presented in previeous works.
Automatic channel selection using shuffled frog leaping algorithm for EEG bas...TELKOMNIKA JOURNAL
Drug addiction is a complex neurobiological disorder that necessitates comprehensive treatment of both the body and mind. It is categorized as a brain disorder due to its impact on the brain. Various methods such as electroencephalography (EEG), functional magnetic resonance imaging (FMRI), and magnetoencephalography (MEG) can capture brain activities and structures. EEG signals provide valuable insights into neurological disorders, including drug addiction. Accurate classification of drug addiction from EEG signals relies on appropriate features and channel selection. Choosing the right EEG channels is essential to reduce computational costs and mitigate the risk of overfitting associated with using all available channels. To address the challenge of optimal channel selection in addiction detection from EEG signals, this work employs the shuffled frog leaping algorithm (SFLA). SFLA facilitates the selection of appropriate channels, leading to improved accuracy. Wavelet features extracted from the selected input channel signals are then analyzed using various machine learning classifiers to detect addiction. Experimental results indicate that after selecting features from the appropriate channels, classification accuracy significantly increased across all classifiers. Particularly, the multi-layer perceptron (MLP) classifier combined with SFLA demonstrated a remarkable accuracy improvement of 15.78% while reducing time complexity.
Explore the innovative world of trenchless pipe repair with our comprehensive guide, "The Benefits and Techniques of Trenchless Pipe Repair." This document delves into the modern methods of repairing underground pipes without the need for extensive excavation, highlighting the numerous advantages and the latest techniques used in the industry.
Learn about the cost savings, reduced environmental impact, and minimal disruption associated with trenchless technology. Discover detailed explanations of popular techniques such as pipe bursting, cured-in-place pipe (CIPP) lining, and directional drilling. Understand how these methods can be applied to various types of infrastructure, from residential plumbing to large-scale municipal systems.
Ideal for homeowners, contractors, engineers, and anyone interested in modern plumbing solutions, this guide provides valuable insights into why trenchless pipe repair is becoming the preferred choice for pipe rehabilitation. Stay informed about the latest advancements and best practices in the field.
Vaccine management system project report documentation..pdfKamal Acharya
The Division of Vaccine and Immunization is facing increasing difficulty monitoring vaccines and other commodities distribution once they have been distributed from the national stores. With the introduction of new vaccines, more challenges have been anticipated with this additions posing serious threat to the already over strained vaccine supply chain system in Kenya.
Courier management system project report.pdfKamal Acharya
It is now-a-days very important for the people to send or receive articles like imported furniture, electronic items, gifts, business goods and the like. People depend vastly on different transport systems which mostly use the manual way of receiving and delivering the articles. There is no way to track the articles till they are received and there is no way to let the customer know what happened in transit, once he booked some articles. In such a situation, we need a system which completely computerizes the cargo activities including time to time tracking of the articles sent. This need is fulfilled by Courier Management System software which is online software for the cargo management people that enables them to receive the goods from a source and send them to a required destination and track their status from time to time.
Automobile Management System Project Report.pdfKamal Acharya
The proposed project is developed to manage the automobile in the automobile dealer company. The main module in this project is login, automobile management, customer management, sales, complaints and reports. The first module is the login. The automobile showroom owner should login to the project for usage. The username and password are verified and if it is correct, next form opens. If the username and password are not correct, it shows the error message.
When a customer search for a automobile, if the automobile is available, they will be taken to a page that shows the details of the automobile including automobile name, automobile ID, quantity, price etc. “Automobile Management System” is useful for maintaining automobiles, customers effectively and hence helps for establishing good relation between customer and automobile organization. It contains various customized modules for effectively maintaining automobiles and stock information accurately and safely.
When the automobile is sold to the customer, stock will be reduced automatically. When a new purchase is made, stock will be increased automatically. While selecting automobiles for sale, the proposed software will automatically check for total number of available stock of that particular item, if the total stock of that particular item is less than 5, software will notify the user to purchase the particular item.
Also when the user tries to sale items which are not in stock, the system will prompt the user that the stock is not enough. Customers of this system can search for a automobile; can purchase a automobile easily by selecting fast. On the other hand the stock of automobiles can be maintained perfectly by the automobile shop manager overcoming the drawbacks of existing system.
CFD Simulation of By-pass Flow in a HRSG module by R&R Consult.pptxR&R Consult
CFD analysis is incredibly effective at solving mysteries and improving the performance of complex systems!
Here's a great example: At a large natural gas-fired power plant, where they use waste heat to generate steam and energy, they were puzzled that their boiler wasn't producing as much steam as expected.
R&R and Tetra Engineering Group Inc. were asked to solve the issue with reduced steam production.
An inspection had shown that a significant amount of hot flue gas was bypassing the boiler tubes, where the heat was supposed to be transferred.
R&R Consult conducted a CFD analysis, which revealed that 6.3% of the flue gas was bypassing the boiler tubes without transferring heat. The analysis also showed that the flue gas was instead being directed along the sides of the boiler and between the modules that were supposed to capture the heat. This was the cause of the reduced performance.
Based on our results, Tetra Engineering installed covering plates to reduce the bypass flow. This improved the boiler's performance and increased electricity production.
It is always satisfying when we can help solve complex challenges like this. Do your systems also need a check-up or optimization? Give us a call!
Work done in cooperation with James Malloy and David Moelling from Tetra Engineering.
More examples of our work https://www.r-r-consult.dk/en/cases-en/
Industrial Training at Shahjalal Fertilizer Company Limited (SFCL)MdTanvirMahtab2
This presentation is about the working procedure of Shahjalal Fertilizer Company Limited (SFCL). A Govt. owned Company of Bangladesh Chemical Industries Corporation under Ministry of Industries.
NO1 Uk best vashikaran specialist in delhi vashikaran baba near me online vas...Amil Baba Dawood bangali
Contact with Dawood Bhai Just call on +92322-6382012 and we'll help you. We'll solve all your problems within 12 to 24 hours and with 101% guarantee and with astrology systematic. If you want to take any personal or professional advice then also you can call us on +92322-6382012 , ONLINE LOVE PROBLEM & Other all types of Daily Life Problem's.Then CALL or WHATSAPP us on +92322-6382012 and Get all these problems solutions here by Amil Baba DAWOOD BANGALI
#vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore#blackmagicformarriage #aamilbaba #kalajadu #kalailam #taweez #wazifaexpert #jadumantar #vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore #blackmagicforlove #blackmagicformarriage #aamilbaba #kalajadu #kalailam #taweez #wazifaexpert #jadumantar #vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore #Amilbabainuk #amilbabainspain #amilbabaindubai #Amilbabainnorway #amilbabainkrachi #amilbabainlahore #amilbabaingujranwalan #amilbabainislamabad
Overview of the fundamental roles in Hydropower generation and the components involved in wider Electrical Engineering.
This paper presents the design and construction of hydroelectric dams from the hydrologist’s survey of the valley before construction, all aspects and involved disciplines, fluid dynamics, structural engineering, generation and mains frequency regulation to the very transmission of power through the network in the United Kingdom.
Author: Robbie Edward Sayers
Collaborators and co editors: Charlie Sims and Connor Healey.
(C) 2024 Robbie E. Sayers
TECHNICAL TRAINING MANUAL GENERAL FAMILIARIZATION COURSEDuvanRamosGarzon1
AIRCRAFT GENERAL
The Single Aisle is the most advanced family aircraft in service today, with fly-by-wire flight controls.
The A318, A319, A320 and A321 are twin-engine subsonic medium range aircraft.
The family offers a choice of engines
Sachpazis:Terzaghi Bearing Capacity Estimation in simple terms with Calculati...Dr.Costas Sachpazis
Terzaghi's soil bearing capacity theory, developed by Karl Terzaghi, is a fundamental principle in geotechnical engineering used to determine the bearing capacity of shallow foundations. This theory provides a method to calculate the ultimate bearing capacity of soil, which is the maximum load per unit area that the soil can support without undergoing shear failure. The Calculation HTML Code included.
A comparative study of mango fruit pest and disease recognition
1. TELKOMNIKA Telecommunication Computing Electronics and Control
Vol. 20, No. 6, December 2022, pp. 1264~1275
ISSN: 1693-6930, DOI: 10.12928/TELKOMNIKA.v20i6.21783 1264
Journal homepage: http://telkomnika.uad.ac.id
A comparative study of mango fruit pest and disease
recognition
Kusrini1
, Suputa2
, Arief Setyanto1
, I Made Artha Agastya1
, Herlambang Priantoro3
, Sofyan Pariyasto1
1
Magister of Informatics Engineering, Universitas AMIKOM Yogyakarta, Yogyakarta, Indonesia
2
Department of Plant Protection, Faculty of Agriculture, Universitas Gadjah Mada, Yogyakarta, Indonesia
3
PT Bank Mandiri, Jakarta, Indonesia
Article Info ABSTRACT
Article history:
Received Sep 17, 2021
Revised Sep 12, 2022
Accepted Sep 22, 2022
Mango is a popular fruit for local consumption and export commodity.
Currently, Indonesian mango export at 37.8 M accounted for 0.115% of
world consumption. Pest and disease are the common enemies of mango that
degrade the quality of mango yield. Specialized treatment in export
destinations such as gamma-ray in Australia, or hot water treatment in
Korea, demands pest-free and high-quality products. Artificial intelligence
helps to improve mango pest and disease control. This paper compares the
deep learning model on mango fruit pests and disease recognition. This
research compares Visual Geometry Group 16 (VGG16), residual neural
network 50 (ResNet50), InceptionResNet-V2, Inception-V3, and DenseNet
architectures to identify pests and diseases on mango fruit. We implement
transfer learning, adopt all pre-trained weight parameters from all those
architectures, and replace the final layer to adjust the output. All the
architectures are re-train and validated using our dataset. The tropical mango
dataset is collected and labeled by a subject matter expert. The VGG16
model achieves the top validation and testing accuracy at 89% and 90%,
respectively. VGG16 is the shallowest model, with 16 layers; therefore, the
model was the smallest size. The testing time is superior to the rest of the
experiment at 2 seconds for 130 testing images.
Keywords:
CNN
InceptionResNet-V2
Inception-V3
Mango pest and disease
ResNet50
VGG16
This is an open access article under the CC BY-SA license.
Corresponding Author:
Kusrini
Magister of Informatics Engineering, Universitas AMIKOM Yogyakarta
Jl. Ringroad Utara Condong Catur Depok Sleman, Yogyakarta, Indonesia
Email: kusrini@amikom.ac.id
1. INTRODUCTION
Mango is a potential commodity for local consumption and export. Indonesian coastal areas with
high sun exposure all year round are suitable for mango. The regional consumption is 0.5 kg per capita per
year, while exports account for 0.11% of world mango consumption. The international demand is high;
however, due to the importer country requirements, Indonesian mango is challenging to enter the global
market. Pest and disease-free requirements in Japan, Australia, and the Korean market hinder Indonesian
mango from accessing their market. Therefore, pest and disease control to ensure the fruit product’s quality
plays a significant role in improving international market acceptance.
The ability of farmers to identify pests and diseases and proper handling is a significant factor.
There are two mango farm models: the big professional farm and household mango trees. On big farms such
as in east java and west java, experienced farmers manage large-scale areas. While in the household mango,
some people grow a few trees around their house. Skilled people carry out pest and disease control on a big
farm with sufficient knowledge. However, it is unavailable for general people with few mango trees around
2. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1265
their houses. An innovative way to disseminate pests and diseases control techniques is desirable to
overcome the knowledge gap.
Mobile application and image recognition are an opportunity to alleviate the pest and disease control
dissemination knowledge problem. The penetration of smartphones is currently at about 56% of Indonesian
citizens. Therefore, it is feasible to deliver knowledge through the smartphone. Image recognition has
matured to detect many visual clues, including the pest and diseases on leaves and fruits. With the help of the
mobile application, image recognition has been implemented in many recognition work for pests and diseases
of plants such as [1], [2].
Deep learning has enjoyed tremendous success in classification tasks, particularly for image data.
The availability of huge labeled datasets such as ImageNet [3] enables researchers to propose, test and
validate many convolutional neural network (CNN)-based architectures. The transfer learning concept
enables researchers to use knowledge learned by other problem sets (datasets) to be implemented in their
specific problems with smaller dataset sizes [4]. In transfer learning concepts, we can use the weight
parameter. To adjust the network to the new problems, fine-tuning was carried out. The adjustment can be
applied to the entire network of only selected layers. Identify the best performance deep learning
architectures, selecting which part of the layer needs tuning/adjusting. Currently, researchers proposed many
well-known deep learning architectures, to name a few: AlexNet [5], AlexNetOWTBn [6], GoogLeNet [7],
Overfeat [8], visual geometry group network (VGGnet) [9], residual neural network (ResNet) [10],
InceptionResnet-V2 [11], Inception-V1 [7], Inception-V2 [12], Inception-V3 [12], and Inception-V4 [11],
and DenseNet [13]. They all have different network structures, number of layers, size of filters and many
other differences. Those lead to different weight parameters ranging from thousands to hundreds of millions.
Consequently, they have different computational complexity, training and testing time.
This research is carried out to extend our previous work on recognizing mango pests on leaves.
The recognition system for pests on mango leaves has been implemented on a mobile application [14].
Following up on suggestions on evaluation results [15], we improve the capability as an extension to the
fruit. This research collects the mango fruit dataset and involves pests and diseases expert to manually
classify the images into five classes. The appearance of pests and diseases of mango fruit in each class is
shown in Figure 1(a) to Figure 1(e).
With the dataset size on hand we expect to be able to recognize the image collected in the real farm
through the mobile application. Accurate and high speed recognition is desired to serve the mobile
application. We aim to seek the most acceptable performance deep learning architecture with high accuracy
and fast recognition. Therefore, we compare available architectures in transfer learning mode and compare
their speed and accuracy.
(a) (b) (c) (d) (e)
Figure 1. Type of mango fruit pest and disease: (a) Capnodium mangiferae, (b) Cynopterus titthaecheilus,
(c) Deanolis albizonalis, (d) Pseudaulacaspis cockerelli, and (e) Pseudococcus longispinus
2. RELATED WORKS
Plant pest and disease recognition based on visual data has attracted computer vision researchers in the
past five decades. Researchers employ a support vector machine (SVM) to classify pomegranate fruit images
into four classes consisting of normal and three infection stages [16]. Support vector machine classifies the
pomegranate images using some features, which are color coherence vector (CCV), color histogram, and shape.
They achieve 82% accuracy on the testing dataset. A classification of 82 disease classes on 12 plants has been
reported by Barbedo et al. [17]. Evaluation of each plant has been carried out independently. The input image
was segmented using guided active contour (GAC). Histogram similarity to the reference image is ranked as the
basis of disease recognition. In [18], the image is segmented into the uninfected and infected regions.
Researchers rely on the hue and co-occurrence matrix of infected leaf images to extract the features. Their work
achieves 95.71% accuracy using SVM to predict the type of leaf disease. In [19], a comparison between a
sparse representation-based classification (SRC), SVM, and artificial neural network (ANN) is carried out to
classify cucumber leaf disease. The SRC outperforms SVM and ANN at 85.7% accuracy. A combination of
3. ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 20, No. 6, December 2022: 1264-1275
1266
dictionary learning and spare representation has been reported 92.9% accuracy on Caltex Leaves dataset [20].
Tan et al. [21] employ SVM to recognize a particular cacao fruit disease called cacao black pod rot (BPR)
using k-means clustering and SVM. Their experiments reported 84% accuracy in recognizing BPR.
In the last decade, deep learning gained popularity and showed a breakthrough performance in
image classification. Computer vision researchers made use of deep learning algorithms in recognizing plant
pests and diseases such as in [22], [23] to recognize various leaf diseases [24] in a controlled condition.
The promising result has been achieved at 96.3% of precision and 99.35% of accuracy. CNN was utilized by
Krizhevsky et al. [5]. The transfer learning concept allows researchers to adopt a model trained by a huge
dataset like image-net and adapt to particular cases with a smaller number of data. Deng et al. [25] implemented
transfer learning and carried out fine-tuning to achieve high-performance classifier. Lu et al. [26] classified rice
leaf disease using CNN and reported testing accuracy at 95.48%. They also identified that the stochastic pooling
layer gave the best results after evaluating three different pooling layers. Wheat is an important food source.
Therefore, in 2017 researchers collected seven classes of wheat disease in the wild called wheat disease
database 2017 (WDD2017). In [27], they reported CNN-based networks with no fully connected layer (FCL)
layers have been superior compared to the original CNN in classifying the WDD2017. Too et al. [28] reported a
comparison of some well-known CNN architectures in classifying the PlantVillage dataset [24]. They evaluated
VGGnet [9], Inception V4 [11], DenseNet [13], and ResNet [10]. According to their experiments, the DenseNet
outperforms the rest of the architectures at 99.75% of accuracy. Researchers introduced a mobile-based wheat
leaf disease recognition at [2]. They used ResNet architecture with 50 layers to carry out a classification task,
and it showed promising classification performance at 96% of accuracy. Ferentinos [29] has put their effort in
classifying leaf disease problems using five CNN architectures which are AlexNet [5], AlexNetOWTBn [6],
GoogLeNet [7], Overfeat [8], VGGnet [9]. They reported that VGGnet outperforms the rest of the
architectures, and it reaches 99.48% accuracy. A study of key factors impacting deep learning performance has
been reported in [30]. They found that image background, image capture conditions, symptom representation,
covariate shift, symptom segmentation, symptom variations, simultaneous disease, and symptom similarity are
impacting factors to the deep learning performance. In [31], independent processing to each color channel input
was introduced. The result was combined as an input of FCL. They evaluated the three channel CNN,
GoogleNet, and LeNet-5 [32] to classify cucumber leaf disease and found that the three channels CNN achieved
the best accuracy at 91.15%. In [33], the apple trunk disease recognition was carried out using VGGnet. They
compared a VGGNet with Focal loss and softmax loss function. The VGGnet using focal loss function better
performance with 2% margin at 94.5% accuracy compared VGGnet with softmax loss function. In [34],
VGGnet was used to recognize mildew diseases and reach 95% of accuracy. Barbedo [35] reported that a
classification task of 14 leaf diseases attain 94% of accuracy on implementation of GoogleNet architecture.
Despite the popularity of mango, there are a limited number of studies on mango pest and diseases
recognition. The author reported 48.95% of accuracy on a recognition task of four diseases and a normal leaf
using SVM [36]. They extract several features from the gray-level co-occurrence matrix (GLCM) matrix
such as contrast, correlation, energy, homogeneity, mean, standard-deviation, entropy, root mean square
(RMS), variance, smoothness, kurtosis, and skewness. Singh et al. [37] used CNN to recognize anthracnose
disease and reported 97.13% accuracy. The obvious visual cue was responsible for the high achievement of
this task. In our previous work, we classified mango pests [1] on affected leaf images. We collected the
dataset [38] from mango farms in Indonesia and organized them into sixteen classes. We implemented
augmentation techniques such as noise addition, blur, contrast, and affine transformation (i.e., rotation and
translation in Cartesian coordinate) in order to improve the performance of VGGnet classifier. According to
our experiments, augmentation successfully improved the accuracy by a 4% margin after using augmented
images in the training phase.
3. RESEARCH METHOD
This is interdisciplinary research that involves mango pest expert and computer scientists.
The dataset was collected around Indonesia. The image collection is labelled by mango pest expert. Once the
dataset is labeled, data size standardization is carried out.
3.1. Dataset
The dataset consists of 653 labeled mango fruit images with five pests and a disease identified. In the
real case, usually only one pest on particular mango fruit as reflected in the dataset each fruit images have a
single pest label. The dataset is divided into training, testing, and validation set at 60%, 20%, and 20%
respectively as presented in Table 1.
4. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1267
Table 1. Number of images in dataset
Pest/disease name Train Valid Test Total
capnodium_mangiferae 38 13 13 64
cynopterus_titthaecheilus 63 21 21 105
deanolis_albizonalis 86 28 28 142
pseudaulacaspis_cockerelli 159 52 52 263
pseudococcus_longispinus 47 16 16 79
Total 393 130 130 653
3.2. Deep learning image classifier
We implement a convolutional neural network using five well-known architectures. Their names are
VGG16, RestNet50, InceptionResNet-V2, Inception-V3, and DenseNet. They will be discussed in following
sub sections.
3.2.1. VGG16 model
We apply a CNN architecture named VGG16 which was used to win Imagenet competition in 2014.
Figure 2 presents the detail architectures. This research adopts transfer learning methods as weight
initialization. The VGG16 network has already been trained upon ImageNet dataset. So the initial weights of
our network are duplicates of the ImageNet pre-trained model. A replacement of the final layers is carried out in
the original VGG16 architecture. All the layer is freezed to retain the trained weight from the ImageNet, while
the training set is performed to train the last replaced layer only. By limiting the weight for the last layer of the
network, we can speed up the training time without sacrificing the classifier accuracy. Consequently, we only
train the last layer, which is the fully connected layer (FCL) with a softmax function.
Figure 2. VGG16 architectures
The VGG16 is a convolutional neural network with 16 layers, and it is quite heavy computing
complexity in the training process. However, the VGG-16’s trained model execution is fast on the personal
computer because CNN is a parameterized learning. By multiplying the saved weights again with a new
sample, the model can predict its class. The training process is executed in a dedicated deep learning server.
Therefore, computational load is not a significant matter. This research develops a high-accuracy model to
serve client applications that recognize the pest from fruit images.
3.2.2. ResNet50 model
He et al. [10] introduced the ResNet model in their publication, which served as the basis for
The ImageNet large-scale visual recognition challenge (ILSVRC) 2015 and Microsoft Common Objects in
Context (COCO) 2015 classification challenges. Their model was ranked first in ImageNet classification with
an error rate of 3.57%. Multiple non-linear layers’ failure to learn identity mappings and the degradation
problem spurred the development of the deep ResNet.
ResNet is a network-in-network (NIN) architecture that is built on a foundation of numerous stacked
residual units. These leftover units are the network’s building blocks. A collection of residual units serves as
the foundation for the ResNet architecture [10]. Convolution, pooling, and layering are used to create the
residual units. The architecture is comparable to that of the VGG network [9], which consists of 33 filters,
although ResNet is approximately eight times deeper. This is because global average pooling is used instead of
fully connected layers. ResNet was further updated to improve accuracy by changing the residual module to use
identity mappings. As in [10], a ResNet model of 50, 101, and 152 layers were built and loaded with pre-trained
weights from ImageNet. Finally, a bespoke softmax layer was constructed for the purpose of identifying plant
diseases. We use ResNet50 in this research. The architecture is shown in Figure 3.
5. ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 20, No. 6, December 2022: 1264-1275
1268
Figure 3. ResNet50 architecture
3.2.3. InceptionResNet-V2 model
InceptionResnet-V2 [11] is a CNN architecture that is a development of InceptionResnet-V1.
InceptionResnet-V2 is used for the transfer learning classification process and is built based on the inception
architecture by combining residual connections. InceptionResnet-V2 was developed by replacing the filter
part of the Inception process [39], [40].
In Figure 4, there are several stages of classification to detect mango images. The input information
is an image that will be converted into a frame to apply image processing techniques. Then the mango image
is detected to produce important features that are in accordance with the characteristics and special
characteristics of the mango fruit [41]. Inception Resnet-V2 in detecting mangoes is used as a high-level
feature extractor that provides image content that can help identify pests on mango fruit [42].
Figure 4. InceptionResNet-V2 architecture
3.2.4. Inception-V3 model
Inception has four versions, namely Inception-V1 [7], Inception-V2 [12], Inception-V3 [12], and
Inception-V4 [11]. The inception model uses several filters on the usual layers. The results of several filters
are combined using a concatenated channel before entering the next iteration [43]. There are 48 Layers in
Inception-V3, which is deeper than its predecessor deep convolutional neural network architecture named
Inception-V1 or GoogLeNet [7].
Inception V3 network structure uses the convolution kernel splitting method to split large volume
integrals into small convolutions. For example, the convolution 3×3 is divided into convolutions 3×1 and
1×3. Through the separation method, the number of parameters can be reduced; hence, network training
speed can be accelerated while spatial features can be extracted more effectively [44].
This study uses one of the deep learning neural network models, the Inception-V3 model used in
TensorFlow to extract and classify mango fruit image features [45], [46]. The Inception-V3 model was used
in TensorFlow to develop an image classifier to classify images of pests on mangoes based on three features:
texture, shape, and color. The architecture is shown in Figure 5.
3.2.5. DenseNet model
In their paper, Huang et al. [13] introduced a densely connected convolutional network architecture.
To ensure maximum information flow between layers in the network, all layers are connected directly with
each other in a feed-forward manner. For each layer, the feature-maps of all preceding layers are used as
inputs, and its own feature maps are used as inputs into all subsequent layers. DenseNets alleviate the
problem of the vanishing-gradient problem and has substantially reduced number of parameters [13]. For this
task of plant disease identification, DenseNets model with121 layers as described in [13] was created.
Additionally, the model was loaded with pre-trained weights from ImageNet. Finally, another fully-connected
model with our own customized softmax on the top layer was created.
6. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1269
Figure 5. Inception-V3 architecture
3.3. Fine tuning
This research adopts the transfer learning concept, where all the architecture above has been pre-trained
using imageNet dataset for 1000 target classes. Transfer learning aims to use the knowledge gained during training
in one type of problem is used to train in another related task or domain [4]. In order to adjust the network to fit
our classification problem, the head of the network is replaced so that the number of target class are five classes.
Fine-tuning is a concept of transfer learning. In our research, fine-tuning is carried out to the entire
network in order to re-train all the weight parameters. Fine-tuned learning is started from the initial condition
where the weight parameters are already trained on other problems. With new training set, it is need an
adjustment for all the layers or particular layers. The researcher can select which layer needs to be re-trained
and freeze other particular layers. Even though the training is needed for adjusting the new problem set,
the initial knowledge can significantly cut the learning effort compared to training from scratch [23]. More
importantly, in manual cases it is more accurate compared to models trained from scratch.
In this research, the CNN models were fine-tuned to identify and classify five categories of fruit
disease with pre-trained models on ImageNet dataset. ImageNet dataset is a huge collection of 1.2 Million
labeled images in 1000 categories. The CNN architectures with the new head are re-train with a small
number of mango fruit images.
4. RESULTS AND DISCUSSION
The research aims to identify the deep learning architecture with acceptable performance and high
accuracy. Testing time and testing accuracy are two main considerations in the problem sets as the algorithms
are designed to serve mobile client applications with multiple requests concurrently. Training time is
important, but it was not the main consideration, since the training will only take place in the modelling task.
It is worth mentioning that the server in these experiments is the prototype of the server that we use to serve
the running pest visual recognition mobile application.
4.1. Experiment setup and parameters
The experiment in this research is conducted using computer server with specification:
− Processor: i9 9900K.
− Memory: 64 Gb.
− GPU: NVIDIA TITAN V, Memory 12 Gb, Tensor Cores 640, CUDA Cores 5120.
This research did not optimize the parameter. The only parameters set is the learning rate at 0.00005. The rest
of the parameters is set at default value. Optimization algorithm is using “Adam”.
4.2. Evaluation metrics
The accuracy and loss for training and validation set were recorded. Training time and validation
time for the entire training and validation set were also become this research focus. Finally, the testing
accuracy and average testing time among the algorithm was recorded to consider the most feasible model to
be implemented in the implementation. The execution time might not repeatable to get the exact similar
number in other research due to different experiment settings, however, the comparison between deep
learning models might show a similar trend.
7. ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 20, No. 6, December 2022: 1264-1275
1270
4.2.1. VGG16
In VGG16, the training loss decreased until 50 epochs. However, the validation loss starts to
fluctuate in epoch 5. It shows us that model start to over fit with the data training. The minimum accuracy and
validation loss are 0.0458 and 0.3108, respectively. The validation accuracy starts to be stagnant in epoch 5.
The maximum training and validation accuracy are 0.9821 and 0.90769, respectively. The accuracy and loss
graphic is shown in Figure 6.
Figure 6. VGG16 training and validation accuracy and loss
4.2.2. ResNet50
In Resnet50, the training loss decrease until 50 epochs. However, the validation loss starts to
increase in epoch 8. It shows us that model start to over fit with the data training. The minimum accuracy and
validation loss are 0.0422 and 0.3885, respectively. The validation accuracy starts to be stagnant in epoch 12.
The maximum training and validation accuracy are 0.9821 and 0.8923, respectively. The accuracy and loss
graphic is shown in Figure 7.
4.2.3. InceptionResNet-V2
In InceptionResNet-V2, the training loss decrease until 50 epochs. However, the validation loss
starts to increase in epoch 20. It shows us that model start to over fit with the data training. The minimum
accuracy and validation loss are 0.0537 and 0.4207, respectively. The validation accuracy starts to be stagnant
in epoch 9. The maximum training and validation accuracy are 0.9796 and 0.8846, respectively. The accuracy
and loss graphic is shown in Figure 8.
Figure 7. RestNet50 training and validation accuracy and loss
8. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1271
Figure 8. InceptionResNet-V2 training and validation accuracy and loss
4.2.4. Inception-V3
In Inception-V3, the training loss decrease until 50 epochs. However, the validation loss starts to be
flat in epoch 12. It shows us that model start to over fit with the data training. The minimum accuracy and
validation loss are 0.0531and 0.5618, respectively. The validation accuracy starts to be stagnant in epoch 10.
The maximum training and validation accuracy are 0.9821 and 0.8692, respectively. The accuracy and loss
graphic is shown in Figure 9.
Figure 9. Inception-V3 training and validation accuracy and loss
Figure 10. DensNet training and validation accuracy and loss
9. ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 20, No. 6, December 2022: 1264-1275
1272
4.2.5. DenseNet121
In DenseNet121, the training loss decrease until 50 epochs. However, the validation loss starts to
increase in epoch 12. It shows us that model start to over fit with the data training. The minimum accuracy and
validation loss are 0.0529 and 0.4579, respectively. The validation accuracy starts to be stagnant in epoch 20.
The maximum training and validation accuracy are 0.9847and 0.8846, respectively. The accuracy and loss
graphic is shown in Figure 10.
4.3. Discussion
According to Table 2, the least layer is VGG-16 model as it is only 16 layers. In the parameter size,
the DenseNet121 produced 9.3 million parameters that is the smallest number parameter among five of them.
Consequently, the model size of DenseNet121 is the smallest too that is 113 Mb. In the training accuracy,
the VGG-16, ResNet50, and Densnet121 can achieve 0.9821 accuracies. Although, the smallest training loss
is achieved by ResNet50 model. The best validation accuracy is achieved by VGG-16 and Resnet50.
However, the VGG-16 has a slightly lower validation loss than ResNet50. The lower loss means the model
can better predict the test data. It is proved by the testing accuracy of VGG-16 can overcome all competitor
as it achieved 0.9076. In addition, the training and testing time of VGG-16 is the smallest comparing the
other models. The training and testing times are 141.72 s and 2.15 s, respectively. VGG 16 architectures
shows its superiority compare to the rest in term of testing accuracy and time. VGG16 is the shallowest
architecture with 16 layers. Therefore, the model size is the smallest compared to deeper architectures. Based on
the results, we can confidence that the VGG-16 model is suitable for fruit disease detection.
Table 2. Accuracy and loss of training and execution time on 50 epochs
Model Layers Parameter
size
Model
size
(mb)
Training
accuracy
Training
loss
Validation
accuracy
Validation
loss
Testing
accuracy
Training
time (s)
Testing
time (s)
VGG-16 16 21,138,757 253 0.9821 0.0458 0.8923 0.3108 0.9076 141.72 2.15
ResNet50 50 28,307,845 340 0.9821 0.0422 0.8923 0.3885 0.8923 173.09 15.67
InceptionResNetV2 164 55,911,141 673 0.9796 0.0537 0.8846 0.4207 0.8846 554.12 49.88
InceptionV3 48 23,901,447 287 0.9821 0.0531 0.8692 0.5618 0.8923 178.75 19.46
DenseNet121 121 9,398,341 113 0.9847 0.0529 0.8846 0.4579 0.8923 274.29 30.9
In Kusrini et al. [1] we implemented VGG16 classifier for recognizing the pest on leaf images. Since
image data collection of infected mango leaf is not easy to collect huge number of data, an augmentation of the
original sample was carried out in order to improve the classification performance. We achieved 71% on
testing data for that experiment, it is due to the cluttered background, visual similarities among many
different classes as mentioned by Barbedo [30]. In this paper, we expand the classification task to the mango
fruit dataset and as can be seen in Table 2 the best achievement reported by VGG16 even without data
augmentation implemented. The fact that the fruit dataset is as small as the leaf dataset but the background
much tidy and the similarity between class much obvious. It is lead to well performance among all the evaluated
architectures.
We also found an interesting fact that the deeper architecture cannot improve better accuracy. The fact that the
problem is simple with only five target classes, low interclass similarity and tidy background enable to simpler
network to capture the pattern of the training very well. Longer network lead to overfitting network and it is
indicated by the high training accuracy while lower validation and training accuracy.
With the small dataset, we can achieve 90.76% of accuracy in test set and it would be quite useful if
we brought thus result to the current implementation on mobile application pest detection. It is because the
implementation is not purely automatics but we put the human in the loop. In the future, along with the
implementation of this recognition task and human feedback, we do expect the rich dataset captured from the
field. With more dataset and human in the loop as the user and the crowd labeling expert we expect more
data and we can retrain the classifier and gain better recognition rate.
The model built in this research will be applied in a mobile application. To reduce the possibility of
errors due to different input data, before the process of classifying pests on fruit, the application will identify
whether the image entered is a fruit image or something else. The identification model will use the results of
previous research [47] by adding fruit data as one of the classes.
5. CONCLUSION
VGG 16 can effectively recognize the mango fruit pest with 90% of accuracy and about 0.0165
second recognition time. The speed and the accuracy is acceptable for mobile application pest recognition
system. The rest of the algorithms shows lower accuracy and time-consuming recognition therefore for
current available dataset we conclude that the implementation of VGG16 is acceptable.
10. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1273
ACKNOWLEDGEMENTS
The research activity leading to the publication has been funded by the Ministry of Education, Culture,
Research and Technology of The Republik Indonesia with contract number 311/E4.1/AK.04.PT/2021,
3281.11/LL5/PG/2021, 188/KONTRAK-LPPM/AMIKOM/VII/2021. Author would like to express gratitude to
The Ministry of Farming for its data support. This research has been made possible by endless administrative
support from the research directorate of the Universitas Amikom Yogyakarta and Faculty of Farming
Universitas Gadjah Mada Indonesia. Also, the authors would like to thank NVIDIA Corporation for the
donation of the GPU used in this study.
REFERENCES
[1] K. Kusrini et al., “Data Augmentation for Automated Pest Classification in Mango Farms,” Computers and Electronics in
Agriculture, vol. 179, 2020, doi: 10.1016/j.compag.2020.105842.
[2] A. Picon, A. A. -Gila, M. Seitz, A. O. -Barredo, J. Echazarra, and A. Johannes, “Deep convolutional neural networks for mobile
capture device-based crop disease classification in the wild,” Computers and Electronics in Agriculture, vol. 161, pp. 280–290,
2019, doi: 10.1016/j.compag.2018.04.002.
[3] O. Russakovsky et al., “ImageNet Large Scale Visual Recognition Challenge,” International Journal of Computer Vision,
vol. 115, pp. 211–252, 2015, doi: 10.1007/s11263-015-0816-y.
[4] S. Panigrahi, A. Nanda, and T. Swarnkar, “A Survey on Transfer Learning,” Intelligent and Cloud Computing, vol. 194, pp. 781–789,
2020, doi: 10.1007/978-981-15-5971-6_83.
[5] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,”
Communications of the ACM, vol. 60, no. 6, pp. 84-90, 2017, doi: 10.1145/3065386.
[6] A. Krizhevsky, “One weird trick for parallelizing convolutional neural networks,” ArXiv, pp. 1-7, 2014. [Online]. Available:
https://arxiv.org/pdf/1404.5997.pdf
[7] C. Szegedy et al., “Going deeper with convolutions,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), 2015, pp. 1-9, doi: 10.1109/CVPR.2015.7298594.
[8] P. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y. LeCun, “OverFeat: Integrated Recognition, Localization and
Detection using Convolutional Networks,” ArXiv, 2014. [Online]. Available: https://arxiv.org/pdf/1312.6229.pdf
[9] K. Simonyan, and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” ArXiv, pp. 1–14,
2015. [Online]. Available: https://arxiv.org/pdf/1409.1556.pdf
[10] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), 2016, pp. 770-778, doi: 10.1109/CVPR.2016.90.
[11] C. Szegedy, S. Ioffe, V. Vanhoucke, and A. Alemi, “Inception-v4, Inception-ResNet and the Impact of Residual Connections on
Learning,” in Proc. of the AAAI Conference on Artificial Intelligence, 2017, vol. 31, no. 1, doi: 10.1609/aaai.v31i1.11231.
[12] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the Inception Architecture for Computer Vision,” 2016
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 2818-2826, doi: 10.1109/CVPR.2016.308.
[13] G. Huang, Z. Liu, L. V. D. Maaten, and K. Q. Weinberger, “Densely Connected Convolutional Networks,” in 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 2261-2269, doi: 10.1109/CVPR.2017.243.
[14] K. Kusrini et al., “Automatic Mango Leaf and Trunk Detection as Supporting Tool of Mango Pest Identifier (MPI),” 2021 3rd
International Conference on Cybernetics and Intelligent System (ICORIS), 2021, pp. 1-7, doi: 10.1109/ICORIS52787.2021.9649558.
[15] Kusrini et al., “User Evaluation of Mobile based Smart Mango Pest Identification,” in 2020 8th International Conference on
Cyber and IT Service Management (CITSM), 2020, pp. 1-7, doi: 10.1109/CITSM50537.2020.9268883.
[16] M. Bhange and H. A. Hingoliwala, “Smart Farming: Pomegranate Disease Detection Using Image Processing,” Procedia
Computer Science, vol. 58, pp. 280–288, 2015, doi: 10.1016/j.procs.2015.08.022.
[17] J. G. A. Barbedo, L. V. Koenigkan, and T. T. Santos, “Identifying multiple plant diseases using digital image processing,”
Biosystems Engineering, vol. 147, pp. 104–116, 2016, doi: 10.1016/j.biosystemseng.2016.03.012.
[18] V. Singh and A. K. Misra, “Detection of plant leaf diseases using image segmentation and soft computing techniques,”
Information Processing in Agriculture, vol. 4, no. 1, pp. 41–49, 2017, doi: 10.1016/j.inpa.2016.10.005.
[19] S. Zhang, X. Wu, Z. You, and L. Zhang, “Leaf image based cucumber disease recognition using sparse representation
classification,” Computers and Electronics in Agriculture, vol. 134, pp. 135–141, 2017, doi: 10.1016/j.compag.2017.01.014.
[20] S. Zeng, B. Zhang, and Y. Du, “Joint distances by sparse representation and locality-constrained dictionary learning for robust
leaf recognition,” Computers and Electronics in Agriculture, vol. 142, pp. 563–571, 2017, doi: 10.1016/j.compag.2017.11.013.
[21] D. S. Tan et al., “AuToDiDAC: Automated Tool for Disease Detection and Assessment for Cacao Black Pod Rot,” Crop
Protection, vol. 103, pp. 98–102, 2018, doi: 10.1016/j.cropro.2017.09.017.
[22] S. Sladojevic, M. Arsenovic, A. Anderla, D. Culibrk, and D. Stefanovic “Deep Neural Networks Based Recognition of Plant
Diseases by Leaf Image Classification,” Computational Intelligence and Neuroscience, 2016, doi: doi: 10.1155/2016/3289801.
[23] S. P. Mohanty, D. P. Hughes, and M. Salathé, “Using deep learning for image-based plant disease detection,” Frontiers in Plant
Science, vol. 7, 2016, doi: 10.3389/fpls.2016.01419.
[24] D. P. Hughes and M. Salathe, “An open access repository of images on plant health to enable the development of mobile disease
diagnostics,” ArXiv, 2015. [Online]. Available: https://arxiv.org/ftp/arxiv/papers/1511/1511.08060.pdf
[25] J. Deng, W. Dong, R. Socher, L. -J. Li, K. Li, and Li F. -Fei, “ImageNet: A large-scale hierarchical image database,” in 2009
IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255, doi: 10.1109/CVPR.2009.5206848.
[26] Y. Lu, S. Yi, N. Zeng, Y. Liu, and Y. Zhang, “Identification of rice diseases using deep convolutional neural networks,”
Neurocomputing, vol. 267, pp. 378–384, 2017, doi: 10.1016/j.neucom.2017.06.023.
[27] J. Lu, J. Hu, G. Zhao, F. Mei, and C. Zhang, “An in-field automatic wheat disease diagnosis system,” Computers and Electronics
in Agriculture¸vol. 142, pp. 369–379, 2017, doi: 10.1016/j.compag.2017.09.012.
[28] E. C. Too, L. Yujian, S. Njuki, and L. Yingchun, “A comparative study of fine-tuning deep learning models for plant disease
identification,” Computers and Electronics in Agriculture, vol. 161, pp. 272–279, 2019, doi: 10.1016/j.compag.2018.03.032.
[29] K. P. Ferentinos, “Deep learning models for plant disease detection and diagnosis,” Computers and Electronics in Agriculture,
vol. 145, pp. 311–318, 2018, doi: 10.1016/j.compag.2018.01.009.
[30] J. G. A. Barbedo, “Factors influencing the use of deep learning for plant disease recognition,” Biosystems Engineering, vol. 172,
pp. 84–91, 2018, doi: 10.1016/j.biosystemseng.2018.05.013.
11. ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 20, No. 6, December 2022: 1264-1275
1274
[31] S. Zhang, W. Huang, and C. Zhang, “Three-channel convolutional neural networks for vegetable leaf disease recognition,”
Cognitive Systems Research, vol. 53, pp. 31–41, 2019, doi: 10.1016/j.cogsys.2018.04.006.
[32] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based Learning Applied to Document Recognition,” in Proc. of the
IEEE, vol. 86, no. 11, pp. 2278-2324, 1998, doi: 10.1109/5.726791.
[33] J. Hang, D. Zhang, P. Chen, J. Zhang, and B. Wang, Identification of Apple Tree Trunk Diseases Based on Improved
Convolutional Neural Network with Fused Loss Functions,” in ICIC 2019: Intelligent Computing Theories and Application, 2019,
pp. 274-283, doi: 10.1007/978-3-030-26763-6_26.
[34] S. Coulibaly, B. K. -Foguem, D. Kamissoko, and D. Traore, “Deep neural networks with transfer learning in millet crop images,”
Computers in Industry, vol. 108, pp. 115–120, 2019, doi: 10.1016/j.compind.2019.02.003.
[35] J. G. A. Barbedo, “Plant disease identification from individual lesions and spots using deep learning,” Biosystems Engineering,
vol. 180, pp. 96–107, 2019, doi: 10.1016/j.biosystemseng.2019.02.002.
[36] K. Srunitha and D. Bharathi, “Mango Leaf Unhealthy Region Detection and Classification,” in Computational Vision and Bio
Inspired Computing, 2018, pp. 422–436, doi: 10.1007/978-3-319-71767-8_35.
[37] U. P. Singh, S. S. Chouhan, S. Jain, and S. Jain, “Multilayer Convolution Neural Network for the Classification of Mango Leaves
Infected by Anthracnose Disease,” IEEE Access, vol. 7, pp. 43721–43729, 2019, doi: 10.1109/ACCESS.2019.2907383.
[38] K. Kusrini et al., “Dataset for pest classification in Mango farms from Indonesia,” Mendeley Data, 2020. [Online]. Available:
https://data.mendeley.com/datasets/94jf97jzc8/1 (accessed Jan. 21, 2021).
[39] C. A. Ferreira et al., “Classification of Breast Cancer Histology Images Through Transfer Learning Using a Pre-trained Inception
Resnet V2,” in ICIAR 2018: Image Analysis and Recognition), 2018, pp. 763–770, doi: 10.1007/978-3-319-93000-8_86.
[40] N. Dominic, Daniel, T. W. Cenggoro, A. Budiarto, and B. Pardamean, “Transfer learning using inception-resnet-v2 model to the
augmented neuroimages data for autism spectrum disorder classification,” Communications in Mathematical Biology and
Neuroscience, pp. 1–21, 2021, doi: 10.28919/cmbn/5565.
[41] A. Thomas, Harikrishnan P. M., Palanisamy P., and V. P. Gopi, “Moving Vehicle Candidate Recognition and Classification Using
Inception-ResNet-v2,” in 2020 IEEE 44th Annual Computers, Software, and Applications Conference (COMPSAC), 2020,
pp. 467-472, doi: 10.1109/COMPSAC48688.2020.0-207.
[42] F. Baldassarre, D. G. Morín, and L. R. -Guirao, “Deep Koalarization: Image Colorization using CNNs and Inception-ResNet-v2,”
ArXiv, pp. 1–12, 2017. [Online]. Available: https://arxiv.org/pdf/1712.03400.pdf
[43] H. V. Hoang and M. T. Tran, “DeepSense-Inception: Gait Identification from Inertial Sensors with Inception-like Architecture
and Recurrent Network,” in 2017 13th International Conference on Computational Intelligence and Security (CIS), 2017,
pp. 594-598, doi: 10.1109/CIS.2017.00138.
[44] N. Dong, L. Zhao, C. H. Wu, and J. F. Chang, “Inception v3 based cervical cell classification combined with artificially extracted
features,” Applied Soft Computing, vol. 93, 2020, doi: 10.1016/j.asoc.2020.106311.
[45] X. Xia, C. Xu, and B. Nan, “Inception-v3 for flower classification,” in 2017 2nd International Conference on Image, Vision and
Computing (ICIVC), 2017, pp. 783-787, doi: 10.1109/ICIVC.2017.7984661.
[46] O. Albatayneh, L. Forslöf, and K. Ksaibati, “Image Retraining Using TensorFlow Implementation of the Pretrained Inception-v3
Model for Evaluating Gravel Road Dust,” Journal of Infrastructure Systems¸ vol. 26, no. 2, 2020, doi: 10.1061/(ASCE)IS.1943-
555X.0000545.
[47] K. Kusrini et al., “Automatic Mango Leaf and Trunk Detection as Supporting Tool of Mango Pest Identifier (MPI),” 2021 3rd
International Conference on Cybernetics and Intelligent System (ICORIS), 2021, pp. 1-7, doi: 10.1109/ICORIS52787.2021.9649558.
BIOGRAPHIES OF AUTHORS
Kusrini is a professor from Universitas AMIKOM Yogyakarta Indonesia. She
finished her doctoral program from Universitas Gadjah Mada Yogyakarta Indonesia in 2010.
She is interested in exploring many things about machine learning and other artificial
intelligence field. She also loves in doing research about decision support system and database.
She is member of the IEEE and IEEE Systems, Man, and Cybernetics Society. She can be
contacted at email: kusrini@amikom.ac.id. Website: http://kusrini.com/.
Suputa currently works at the Department of Plant Protection, Faculty of
Agriculture, Universitas Gadjah Mada. He conducts research in the area of insect taxonomy,
systematic and integrated pest management (IPM), generally in Agricultural Entomology. His
expertise includes the identification of fruit flies, ants, aquatic insects, other insect pests using
morphological and molecular characteristics (DNA) and also consultants on integrated pest
management practices. His current research on Area Wide IPM of fruit fly. He can be
contacted at email: puta@ugm.ac.id. Website: https://hpt.faperta.ugm.ac.id/suputa/.
12. TELKOMNIKA Telecommun Comput El Control
A comparative study of mango fruit pest and disease recognition (Kusrini)
1275
Arief Setyanto was born in 1975 in Banyumas, Central Java, Indonesia. Currently
he works as a lecturer in the Universitas Amikom Yogyakarta Indonesia. He received his PhD
in The School of Computer Science and Electronics Engineering (CSEE), University of Essex,
Colchester The United Kingdom in 2016 under Indonesian Government scholarship. He
received bachelor and Master degree from Gadjah Mada University, Yogyakarta, Indonesia in
1998, 2003 respectively. His research interest includes, video/image segmentation and
understanding, object tracking, visual content metadata, information retrieval and
cryptography. He can be contacted at email: arief_s@amikom.ac.id.
I Made Artha Agastya earned a Master’s degree in information technology,
focusing on pattern recognition and data clustering, from Universitas Gadjah Mada, Indonesia
in 2017. He is presently enrolled at Taylor’s University, Malaysia, where he is pursuing a PhD
in signal processing and recognition. Also, he is a lecturer in the informatics department,
Universitas AMIKOM Yogyakarta, Indonesia. He was involved in two grant projects funded
by the Ministry of Education, Culture, Research and Technology, Indonesia. Among his
research interests are machine learning, affective computing, EEG signal processing, deep
learning, and computer vision. He can be contacted at email: artha.agastya@amikom.ac.id.
Herlambang Priantoro is a Technical Analyst at PT Bank Mandiri (Persero)
Tbk. Experienced Development Team with a demonstrated history of working in the
information technology and services industry. He can be contacted at email:
lambanx@gmail.com.
Sofyan Pariyasto received the M. Kom degree in Amikom University
Yogyakarta in 2019. He was born in Sragen in 1989. Currently, he is active as a Product
Owner in a growing private company. He has interests in AI, programming and data
processing. He can be contacted at email: sofyanpariyasto@gmail.com.