SlideShare a Scribd company logo
P R E S E N T E D B Y
PAT WA R I K E TA K I J .
2 0 1 7 M N S 0 1 0
U N D E R T H E G U I D A N C E
O F
D R . M R S . WA G H M A R E J . M .
Contents
 Introduction
 ConvNet Layers
 Activation functions
 Applications of CNN
 Image classification
 Traffic sign recognition
 Vehicle detection
 Some another interesting application
 Conclusion
 References
Introduction
 What is ANN?
 A neural network is a system of interconnected artificial “neurons” that exchange
messages between each other.
fig2. Training of neural network
Fig3. Illustration of neuron and mathematical model
 What is Convolutional neural network?
 A CNN is a special case of the ANN . A CNN consists of one or more convolutional
layers, often with a sub-sampling layer, which are followed by one or more fully
connected layers as in a standard neural network.
 convolution layers play the role of feature extractor
Fig 4 Typical block diagram of a CNN
ConvNet Layers
1) Convolution layers
 The convolution operation extracts different features of the input. The first
convolution layer extracts low-level features like edges, lines, and corners.
2) Pooling Layer/subsampling layer
 3) Non-linear layers
 Neural networks in general and CNNs in particular rely on a non-linear “trigger”
function to signal distinct identification of likely features on each hidden layer.
CNNs may use a variety of specific functions such as rectified linear units (ReLUs)
and continuous trigger (non-linear) functions to efficiently implement this non-linear
triggering.
 ReLU(Rectified Linear Unit)
implements function y = max(x,0).
 Continuous trigger (non-linear) function
 The non-linear layer operates element by element in each feature. A
continuous trigger function can be hyperbolic tangent (Figure I), absolute of
hyperbolic tangent (Figure II), or sigmoid (Figure III).
Figure I Figure II Figure III
Figure IV : tanh processing
4) Fully connected layers
Activation Functions
 Activation functions are non-linarites that take on a single number and do some
mathematical operations on it.
1. Sigmoid:
 This non-linearity takes as input a real-valued function and outputs value in the
range of 0 and 1.
2.Tanh:
 Tanh can be considered as a scaled up version of sigmoid, outputting values in the
range of -1 and 1.
tanh (x) = 2f (2 x)−1
3. Leaky ReLU:
 Slight modification to ReLU resulted in Leaky ReLU given by Equation.
Applications of CNN
 The main categories are as follows:
1. image and pattern recognition
2. speech recognition
3. natural language processing
4. video analysis
Image classification
 Adaboost can effectively control the quantity of weak classifiers, which greatly
increases execution efficiency through integration; therefore, it has also been applied
to real-time object testing.
 This effectively reduces parameter redundancy, which results in greater
generalization and higher training efficiency.
 The first step in this paper is to introduce the basic knowledge of the convolution,
pooling, weight decay, and dropout of the convolution neural network, as well as the
theories and knowledge of ensemble learning.
 With the convolution neural network and the learning principles of Adaboost, this
paper takes the initiative to propose the Boost Convolutional Neural Network, and
conducts experimentation and analysis of image classification with the benchmark
datasets of CIFAR-10 image classification.
 BOOSTING ENSEMBLE LEARNING
 As shown in Figure below , ensemble learning completes learning tasks by constructing and
cooperating with several learners, meaning the so-called ‘‘collectivism’’, where a group of
individual learners are trained, and a strategy is then adopted to combine them.
 Boosting Algorithm
 Boosting is an algorithm concept that combines weak classifiers through certain
methods to form a strong classifier with high classification performance.
 Process of boosting classification:
1. Train the first weak classifier (h1) in the initialized training dataset
2. Combine the poorly-trained data of the weak classifier (h1) with new data to form
the training data for a new round, and train the second weak classifier (h2)
3. Combine the poorly-trained data of weak classifier (h1) and weak classifier (h2)
with new data to form the training data for a new round, and train the third weak
classifier (h3)
4. Repeat Steps 2-3 until an adequate number of weak classifiers have been attained.
5. Form a strong classifier through the weighted voting of all weak classifiers.
 ADABOOST ALGORITHM(Adaptive Boosting)
As its name suggests, according to the trained classifiers, AdaBoost can self-adjust
weak classifiers after learning and is sensitive to noise data and outliers. In some
tasks, it can efficiently resist overfitting.
The concept of Adaboost can be described, as follows:
 the given training dataset T = {(x1, y1), (x2, y2) . . . (xN, yN)}, where x ∈ X, X ∈ R
n , and yi is a mark set {−1, +1}.
 Adaboost aims to acquire a series of weak classifiers from the training data, and
then, combines these weak classifiers into a strong classifier. Moreover, its
computational process can be described, as follows:
1. The weight distribution of the initial training data; each training sample is given the
same initial weight: 1/N i.e
D1 = (w11,w12 . . . w1i . . . ,w1N ), w1i = 1/ N
i = 1, 2, . . . ,N
2. In the case of m = 1, 2, . . . , M uses the training data with the weight distribution
(Dm) in learning to obtain a basic binomial classifier, as follows:
Gm(x) : x → {−1, +1}
3. the rate of a wrong classification of classifier (Gm(x)) in the training dataset is, as
follows:
4. coefficient (αm) of Gm(x)) is calculated, which refers to the voting weight of
Gm(x)) in the final classier:
5. the weight distribution of the updated training datasets is, as follows:
Zm is a normalized factor
6. the linear combination of basic classifiers is established:
The final classifier is then obtained is:
Proposed BoostCNN structure
BoostCNN training procedure.
 SIMULATION EXPERIMENT AND ANALYSIS
 To analyze the performance of the BoostCNN model, this chapter uses the four-
layered convolution network and Softmax classifier as the feature learning
architecture, and adopts the AdaBoost algorithm to generate several Softmax
classifiers as the output.
 CIFAR-10 DATASET
 the CIFAR-10 dataset is an image dataset whose 60,000 colorful images (32 ∗ 32)
are classified into 10 categories. Specifically, 50,000 images are classified as
training data, while 10,000 images are classified as testing data. As shown in Figure
17, the 10 categories are plane, car, bird, cat, deer, dog, frog, horse, ship, and truck,
and each category includes 6,000 images.
Experimental comparison of CIFAR-10 testing datasets.
CONCLUSION
 This study focuses on how to combine a convolution neural network with AdaBoost
to enhance the image identification performance of the learning algorithms.
 After the convolution neural network is trained into the deep feature extraction
model, and the original images are converted to acquire the deep features, AdaBoost
is used for ensemble learning.
 The conclusions of this paper are, as follows:
1) After the feature extraction of the deep convolution neural network, the original
image data are fully abstracted, thus, the traditional learning algorithms can also
effectively learn highly complicated image data.
2) Through the comparative experiments of the CIFAR-10 image dataset, AdaBoost
can generally enhance the performance of base learners by 3%.
Traffic sign recognition
 Traffic sign recognition (TSR) represents an important feature of advanced driver
assistance systems, contributing to the safety of the drivers and vehicles as well.
 Developing TSR systems requires the use of computer vision techniques, which
could be considered fundamental in the field of pattern recognition in general.
Despite all the previous works and research that has been achieved, traffic sign
detection and recognition still remain a very challenging problem, precisely if we
want to provide a real time processing solution.
 In this paper, author present a comparative and analytical study of the two major
approaches for traffic sign detection and recognition.
 The first approach is based on the color segmentation technique and
convolutional neural networks (C-CNN),
 while the second one is based on the fast region-based convolutional neural
networks approach (Fast R-CNN).
 COLOR SEGMENTATION & CNN BASED APPROACH
 The C-CNN method consists of selecting a set of regions for interest (ROIs) by
applying a color tresholding on the input image, thus reducing the search space.
 Then, a trained CNN is used to classify the ROI (whether it contains a traffic sign or
not), followed by another CNN with the same architecture that is used to recognize
the detected traffic signs.
 Using the HSV features and grayscale color spaces, color thresholding can be
applied to a given image or frame.
 After color detection, an additional step is required in order to verify if the
corresponding regions ROIs are a traffic sign or not, by using a CNN model as a
classifier.
 After converting RGB color space into HSV color space the CNN is applied as a
model.
 It is fairly simple and has 7 layers: 3 convolutional layers, 3 maxpooling layers for
feature extraction and one fully connected layer as a classifier. The CNN was trained
to recognize two classes: traffic sign / no traffic sign. For that, we merged two
datasets: the GTSRB dataset and 30,000 random samples of the Cifar-10 dataset
 After classifying the extracted ROIs, we apply a second CNN with same architecture
of the first one, to recognize the detected traffic signs in the ROIs.
 FAST R-CNN BASED APPROACH
 Fast R-CNN was proposed to fix the disadvantages of R-CNN [28] and SPPnet,
while improving their speed and accuracy. The Fast R-CNN method has several
advantages:
1. Higher detection quality than R-CNN, and SPPnet,
2. Training is single-stage, using a multi-task loss.
3. No disk storage is required for feature caching.
4. Training can update all network layers.
 The Fast R-CNN network takes as input an entire image and a set of object
proposals that are calculated by an external algorithm. The network first processes
the whole image with several convolutional (conv) and max pooling layers to
produce a conv feature map. Then, for each object proposal a region of interest
(ROI) pooling layer extracts a fixed-length feature vector from the feature map.
 Each feature vector is fed into a sequence of fully connected(fc) layers that finally
branch into two sibling output layers:
 One that produces softmax probability estimates over K object classes plus a
catch-all “background” class.
 Another layer that outputs four real-value numbers for each of the K object
classes. Each set of 4 values encodes refined bounding-box positions for one of
the K classes.
 EXPERIMENTAL RESULTS
 In this case they used python language for implementation. For color segmentation,
we used the OpenCv API and the well-known deep learning API Keras with
Tensorflow backend were used to train the CNNs models in both approaches.
1. C-CNN APPROACH RESULTS
Here they trained two CNNs with the same architecture, the first one for classifying
the ROIs extracted after applying the color segmentation on input frames and the
second one to recognize the detected traffic signs. The training time took 6 hours on
an Intel core i7 3612QM 8GB 1T.
2. C-CNN APPROACH RESULTS
 With the Fast R-CNN approach, we could achieve 94.8% accuracy on test set (the
GTSDB). The training time took 3 days, on a server with a GTX 1080 TI ROC -
11GO graphic card
 Even though the Fast R-CNN is a very good approach and showed interesting results
in pattern detection & recognition in general, but if the training dataset is not
balanced and big, that limits the performance of the approach.
 CONCLUSION
 This paper presented an analytical study of two effective and efficient road sign
detection and recognition approaches.
 The experimental results achieved after testing both of the methods on the German
Traffic Sign Detection & Recognition datasets, conclude that the Fast R-CNN is so
much faster than the C-CNN method, also it is invariant to illumination changes.
 On the other hand, even though the C-CNN approach is slow and sensitive to
weather conditions, it is invariant to scale and viewing angle.
Vehicle detection
 Vehicle detection and counting in aerial images is important for a wide range of
applications such as urban planning and traffic management.
 Many methods have been introduced in the literature for solving this problem. These
methods are either based on shallow learning or deep learning approaches. However,
these methods suffer from relatively low precision and recall rate.
 This paper introduces an automated vehicle detection and counting system in aerial
images.
 The proposed system utilizes convolution neural network (CNN) to regress a vehicle
spatial density map across the aerial image. It has been evaluated on two publicly
available datasets namely Munich and OIRDS.
 The experimental results show that our proposed system is efficient and effective,
and produces higher precision and recall rate than the comparative methods.
 SHALLOW-LEARNING-BASED METHODS
 The general strategy followed in this group relies on handcrafted features extraction
followed by a classifier or cascade of classifiers.
 authors proposed a system for car counting in aerial images captured by UAV. They
have reduced search space by selecting the regions where cars might exist using a
supervised classifier then extracted feature points using scale invariant feature
transform (SIFT).
 Then support vector machine (SVM) has been used in order to discriminate
between the cars and all other objects. Four steps for car detection system have been
introduced.
1. The proposed system starts with selecting the areas that might have cars.
2. Then, two sets of histogram of oriented gradients (HOG) features are extracted for
vertical and horizontal filtering directions.
3. The discrimination between the cars and other objects has been performed by one
of three suggested techniques: mutual information measure, normalized cross
correlation, and combination of the correlation measure with SVM classification.
4. The discrimination is obtained by associating an orientation value to the points
classified as cars. Finally, the points that belong to the same car are merged.
 DEEP-LEARNING-BASED METHODS
 Most of the works proposed in this category use convolution neural network for
automatic features extraction. In deep convolutional neural network with multi-scale
spatial pyramid pooling (SPP) has been employed in extracting the target patterns
with different sizes. However, input images have been pre-processed by maximum
normed gradient algorithm in order to restore the edges of the objects.
 Another deep learning approach has been introduced. In this , the input image has
been segmented into small homogeneous regions. Then the features in the
segmented regions are extracted using pre-trained convolutional neural network
(CNN) by a sliding-window approach.
 Windows are classified using support vector machine (SVM) into car and no-car
classes. Finally, post-processing is done such as morphological dilation to smooth
the detected regions and fill the holes.
 The proposed system
1. FULLY CONVOLUTIONAL REGRESSION NETWORK(FCRN)
The proposed system. (a) Training phase. (b) Inference Phase.
3. IMPLEMENTATION DETAILS
 The implementation of the proposed architecture is based on Tensorflow
 During training phase, 224x224 random patches were selected from the aerial
image. The selected patch contains at least one vehicle. Thus, patches with no
vehicles were not chosen during training. In order to increase the amount of training
examples, data augmentation techniques were utilized such as rotation, horizontal
and vertical flipping and shifting.
 The mean square error target function is used, as follows:
where, X is the input patch with M samples,
are all trainable parameters,
YP is the predicted density map, and YT
YT is the ground truth annotation.
 RMSprop optimizer has been used for updating the parameter values.
 DATASETS DESCRIPTION
 The proposed system has been evaluated on two public datasets namely DLR
Munich vehicle dataset provided by Remote Sensing Technology Institute of the
German Aerospace Center and Overhead Imagery Research Data Set (OIRDS)
dataset.
 Munich dataset contains 20 images (5616 x 3744 pixels) taken by DLR 3K camera
system at a height of 1000 m above the ground over the area of Munich, Germany.
This dataset contains 3418 cars and 54 trucks annotated in the training image set and
5799 cars and 93 trucks annotated in testing image set.
 OIRDS dataset This dataset contains 907 aerial images with approximately1800
annotated vehicles.
 QUANTITATIVE EVALUATION AND COMPARISON
F1- F1 score
Performance Comparison between the proposed method and the state-of-the-art methods.
 CONCLUSION
 A novel vehicle detection and counting method has been introduced using
convloutional regression neural network. In the proposed system, we have used
regression model in order to predict the density map of the input patches. Then, the
output of FCRN goes under empirical threshold which results a binary image.
 Finally, a simple connected component algorithm is used for finding the locations
and count of the blobs that represent the detected vehicles. The results of the
proposed architecture outperforms the state-of-the-art methods.
 have achieved the highest true positive rate and the lowest false alarm rate.
Some another Interesting applications
 Mixtures of Lightweight Deep Convolutional Neural Networks: Applied to
Agricultural Robotics
 Aircraft Type Recognition Based on Segmentation With Deep Convolutional Neural
Networks
 Deep Convolutional Neural Networks and Data Augmentation for Environmental
Sound Classification
 Fast Deep Neural Networks with Knowledge Guided Training and Predicted
Regions of Interests for Real-time Video Object Detection
Conclusion
 CNNs give the best performance in pattern/image recognition problems and even
outperform humans in certain cases. Cadence has achieved best-in-industry results
using proprietary algorithms and architectures with CNNs.
 Now we have studied CNN and its components along with the applications. Also we
have studied three IEEE papers thoroughly that are having different applications that
are Image classification, traffic sign recognition and Vehicle detection.
 In previous section I have given more interesting applications you may go through
it.
References
[1]Samer Hijazi, Rishi Kumar, and Chris Rowen, IP Group, Cadence,”Using
Convolutional Neural Networks for Image Recognition”.
[2]Neena Aloysius and Geetha M,” A Review on Deep Convolutional Neural
Networks”, International Conference on Communication and Signal Processing,
April 6-8, 2017, India.
[3]Keiron O’Shea1 and Ryan Nash2,” An Introduction to Convolutional Neural
Networks”
[4] Kaoutar Sefrioui Boujemaa, Afaf Bouhoute, Karim Boubouh and Ismail Berrada,”
Traffic sign recognition using convolutional neural networks”, 2017 International
Conference on Wireless Networks and Mobile Communications (WINCOM).
[5] SHIN-JYE LEE, TONGLIN CHEN, LUN YU, AND CHIN-HUI LAI,” Image
Classification Based on the Boost Convolutional Neural Network”, Digital Object
Identifier 10.1109/ACCESS.2018.2796722.
[6] HILAL TAYARA, KIM GIL SOO , AND KIL TO CHONG ,“Vehicle Detection and
Counting in High-Resolution Aerial Images Using Convolutional Regression Neural
Network ”, Digital Object Identifier 10.1109/ACCESS.2017.DOI.
Thank You!

More Related Content

What's hot

Introduction to Deep learning
Introduction to Deep learningIntroduction to Deep learning
Introduction to Deep learning
leopauly
 
Machine Learning - Convolutional Neural Network
Machine Learning - Convolutional Neural NetworkMachine Learning - Convolutional Neural Network
Machine Learning - Convolutional Neural Network
Richard Kuo
 
Convolutional neural network
Convolutional neural networkConvolutional neural network
Convolutional neural network
MojammilHusain
 
Convolutional neural network
Convolutional neural network Convolutional neural network
Convolutional neural network
Yan Xu
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
Overview of Convolutional Neural Networks
Overview of Convolutional Neural NetworksOverview of Convolutional Neural Networks
Overview of Convolutional Neural Networks
ananth
 
Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)
Muhammad Haroon
 
CNN Tutorial
CNN TutorialCNN Tutorial
CNN Tutorial
Sungjoon Choi
 
Image classification using cnn
Image classification using cnnImage classification using cnn
Image classification using cnn
SumeraHangi
 
Deep learning
Deep learningDeep learning
Deep learning
Kuppusamy P
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning
Asma-AH
 
Convolutional neural network from VGG to DenseNet
Convolutional neural network from VGG to DenseNetConvolutional neural network from VGG to DenseNet
Convolutional neural network from VGG to DenseNet
SungminYou
 
Deep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural NetworksDeep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural Networks
Christian Perone
 
Cnn
CnnCnn
Convolutional neural network
Convolutional neural networkConvolutional neural network
Convolutional neural network
Ferdous ahmed
 
Deep learning
Deep learningDeep learning
Deep learning
Ratnakar Pandey
 
Machine Learning - Object Detection and Classification
Machine Learning - Object Detection and ClassificationMachine Learning - Object Detection and Classification
Machine Learning - Object Detection and Classification
Vikas Jain
 
Image classification using CNN
Image classification using CNNImage classification using CNN
Image classification using CNN
Noura Hussein
 
AlexNet
AlexNetAlexNet
AlexNet
Bertil Hatt
 
Neural networks and deep learning
Neural networks and deep learningNeural networks and deep learning
Neural networks and deep learning
Jörgen Sandig
 

What's hot (20)

Introduction to Deep learning
Introduction to Deep learningIntroduction to Deep learning
Introduction to Deep learning
 
Machine Learning - Convolutional Neural Network
Machine Learning - Convolutional Neural NetworkMachine Learning - Convolutional Neural Network
Machine Learning - Convolutional Neural Network
 
Convolutional neural network
Convolutional neural networkConvolutional neural network
Convolutional neural network
 
Convolutional neural network
Convolutional neural network Convolutional neural network
Convolutional neural network
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
 
Overview of Convolutional Neural Networks
Overview of Convolutional Neural NetworksOverview of Convolutional Neural Networks
Overview of Convolutional Neural Networks
 
Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)
 
CNN Tutorial
CNN TutorialCNN Tutorial
CNN Tutorial
 
Image classification using cnn
Image classification using cnnImage classification using cnn
Image classification using cnn
 
Deep learning
Deep learningDeep learning
Deep learning
 
Image Classification using deep learning
Image Classification using deep learning Image Classification using deep learning
Image Classification using deep learning
 
Convolutional neural network from VGG to DenseNet
Convolutional neural network from VGG to DenseNetConvolutional neural network from VGG to DenseNet
Convolutional neural network from VGG to DenseNet
 
Deep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural NetworksDeep Learning - Convolutional Neural Networks
Deep Learning - Convolutional Neural Networks
 
Cnn
CnnCnn
Cnn
 
Convolutional neural network
Convolutional neural networkConvolutional neural network
Convolutional neural network
 
Deep learning
Deep learningDeep learning
Deep learning
 
Machine Learning - Object Detection and Classification
Machine Learning - Object Detection and ClassificationMachine Learning - Object Detection and Classification
Machine Learning - Object Detection and Classification
 
Image classification using CNN
Image classification using CNNImage classification using CNN
Image classification using CNN
 
AlexNet
AlexNetAlexNet
AlexNet
 
Neural networks and deep learning
Neural networks and deep learningNeural networks and deep learning
Neural networks and deep learning
 

Similar to CNN and its applications by ketaki

Human Head Counting and Detection using Convnets
Human Head Counting and Detection using ConvnetsHuman Head Counting and Detection using Convnets
Human Head Counting and Detection using Convnets
rahulmonikasharma
 
Introduction to Convolutional Neural Networks
Introduction to Convolutional Neural NetworksIntroduction to Convolutional Neural Networks
Introduction to Convolutional Neural Networks
ParrotAI
 
28 01-2021-05
28 01-2021-0528 01-2021-05
28 01-2021-05
AdemarAlves7
 
Hand gestures recognition seminar_ppt.pptx.pdf
Hand gestures recognition seminar_ppt.pptx.pdfHand gestures recognition seminar_ppt.pptx.pdf
Hand gestures recognition seminar_ppt.pptx.pdf
SwathiSoman5
 
Image De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural NetworkImage De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural Network
aciijournal
 
Electricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkElectricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkNaren Chandra Kattla
 
Image De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural NetworkImage De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural Network
aciijournal
 
IMAGE DE-NOISING USING DEEP NEURAL NETWORK
IMAGE DE-NOISING USING DEEP NEURAL NETWORKIMAGE DE-NOISING USING DEEP NEURAL NETWORK
IMAGE DE-NOISING USING DEEP NEURAL NETWORK
aciijournal
 
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
csandit
 
Introduction to convolutional networks .pptx
Introduction to convolutional networks .pptxIntroduction to convolutional networks .pptx
Introduction to convolutional networks .pptx
ArunNegi37
 
Web Spam Classification Using Supervised Artificial Neural Network Algorithms
Web Spam Classification Using Supervised Artificial Neural Network AlgorithmsWeb Spam Classification Using Supervised Artificial Neural Network Algorithms
Web Spam Classification Using Supervised Artificial Neural Network Algorithms
aciijournal
 
Face recognition using artificial neural network
Face recognition using artificial neural networkFace recognition using artificial neural network
Face recognition using artificial neural networkSumeet Kakani
 
Artificial Neural Network for machine learning
Artificial Neural Network for machine learningArtificial Neural Network for machine learning
Artificial Neural Network for machine learning
2303oyxxxjdeepak
 
Machine learning Module-2, 6th Semester Elective
Machine learning Module-2, 6th Semester ElectiveMachine learning Module-2, 6th Semester Elective
Machine learning Module-2, 6th Semester Elective
MayuraD1
 
cnn.pdf
cnn.pdfcnn.pdf
cnn.pdf
Amnaalia
 
Vector-Based Back Propagation Algorithm of.pdf
Vector-Based Back Propagation Algorithm of.pdfVector-Based Back Propagation Algorithm of.pdf
Vector-Based Back Propagation Algorithm of.pdf
Nesrine Wagaa
 
Batch normalization presentation
Batch normalization presentationBatch normalization presentation
Batch normalization presentation
Owin Will
 
Mnist report ppt
Mnist report pptMnist report ppt
Mnist report ppt
RaghunandanJairam
 
Mnist report
Mnist reportMnist report
Mnist report
RaghunandanJairam
 

Similar to CNN and its applications by ketaki (20)

Human Head Counting and Detection using Convnets
Human Head Counting and Detection using ConvnetsHuman Head Counting and Detection using Convnets
Human Head Counting and Detection using Convnets
 
Introduction to Convolutional Neural Networks
Introduction to Convolutional Neural NetworksIntroduction to Convolutional Neural Networks
Introduction to Convolutional Neural Networks
 
28 01-2021-05
28 01-2021-0528 01-2021-05
28 01-2021-05
 
Hand gestures recognition seminar_ppt.pptx.pdf
Hand gestures recognition seminar_ppt.pptx.pdfHand gestures recognition seminar_ppt.pptx.pdf
Hand gestures recognition seminar_ppt.pptx.pdf
 
Image De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural NetworkImage De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural Network
 
Electricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkElectricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural Network
 
Image De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural NetworkImage De-Noising Using Deep Neural Network
Image De-Noising Using Deep Neural Network
 
IMAGE DE-NOISING USING DEEP NEURAL NETWORK
IMAGE DE-NOISING USING DEEP NEURAL NETWORKIMAGE DE-NOISING USING DEEP NEURAL NETWORK
IMAGE DE-NOISING USING DEEP NEURAL NETWORK
 
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
Objective Evaluation of a Deep Neural Network Approach for Single-Channel Spe...
 
Introduction to convolutional networks .pptx
Introduction to convolutional networks .pptxIntroduction to convolutional networks .pptx
Introduction to convolutional networks .pptx
 
Web Spam Classification Using Supervised Artificial Neural Network Algorithms
Web Spam Classification Using Supervised Artificial Neural Network AlgorithmsWeb Spam Classification Using Supervised Artificial Neural Network Algorithms
Web Spam Classification Using Supervised Artificial Neural Network Algorithms
 
Face recognition using artificial neural network
Face recognition using artificial neural networkFace recognition using artificial neural network
Face recognition using artificial neural network
 
Artificial Neural Network for machine learning
Artificial Neural Network for machine learningArtificial Neural Network for machine learning
Artificial Neural Network for machine learning
 
N ns 1
N ns 1N ns 1
N ns 1
 
Machine learning Module-2, 6th Semester Elective
Machine learning Module-2, 6th Semester ElectiveMachine learning Module-2, 6th Semester Elective
Machine learning Module-2, 6th Semester Elective
 
cnn.pdf
cnn.pdfcnn.pdf
cnn.pdf
 
Vector-Based Back Propagation Algorithm of.pdf
Vector-Based Back Propagation Algorithm of.pdfVector-Based Back Propagation Algorithm of.pdf
Vector-Based Back Propagation Algorithm of.pdf
 
Batch normalization presentation
Batch normalization presentationBatch normalization presentation
Batch normalization presentation
 
Mnist report ppt
Mnist report pptMnist report ppt
Mnist report ppt
 
Mnist report
Mnist reportMnist report
Mnist report
 

Recently uploaded

Cambridge International AS A Level Biology Coursebook - EBook (MaryFosbery J...
Cambridge International AS  A Level Biology Coursebook - EBook (MaryFosbery J...Cambridge International AS  A Level Biology Coursebook - EBook (MaryFosbery J...
Cambridge International AS A Level Biology Coursebook - EBook (MaryFosbery J...
AzmatAli747758
 
Synthetic Fiber Construction in lab .pptx
Synthetic Fiber Construction in lab .pptxSynthetic Fiber Construction in lab .pptx
Synthetic Fiber Construction in lab .pptx
Pavel ( NSTU)
 
Palestine last event orientationfvgnh .pptx
Palestine last event orientationfvgnh .pptxPalestine last event orientationfvgnh .pptx
Palestine last event orientationfvgnh .pptx
RaedMohamed3
 
The Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official PublicationThe Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official Publication
Delapenabediema
 
Chapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptxChapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptx
Mohd Adib Abd Muin, Senior Lecturer at Universiti Utara Malaysia
 
Additional Benefits for Employee Website.pdf
Additional Benefits for Employee Website.pdfAdditional Benefits for Employee Website.pdf
Additional Benefits for Employee Website.pdf
joachimlavalley1
 
TESDA TM1 REVIEWER FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
TESDA TM1 REVIEWER  FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...TESDA TM1 REVIEWER  FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
TESDA TM1 REVIEWER FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
EugeneSaldivar
 
Instructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptxInstructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptx
Jheel Barad
 
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
siemaillard
 
Digital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and ResearchDigital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and Research
Vikramjit Singh
 
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
siemaillard
 
The Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve ThomasonThe Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve Thomason
Steve Thomason
 
Home assignment II on Spectroscopy 2024 Answers.pdf
Home assignment II on Spectroscopy 2024 Answers.pdfHome assignment II on Spectroscopy 2024 Answers.pdf
Home assignment II on Spectroscopy 2024 Answers.pdf
Tamralipta Mahavidyalaya
 
Fish and Chips - have they had their chips
Fish and Chips - have they had their chipsFish and Chips - have they had their chips
Fish and Chips - have they had their chips
GeoBlogs
 
Unit 2- Research Aptitude (UGC NET Paper I).pdf
Unit 2- Research Aptitude (UGC NET Paper I).pdfUnit 2- Research Aptitude (UGC NET Paper I).pdf
Unit 2- Research Aptitude (UGC NET Paper I).pdf
Thiyagu K
 
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCECLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
BhavyaRajput3
 
ESC Beyond Borders _From EU to You_ InfoPack general.pdf
ESC Beyond Borders _From EU to You_ InfoPack general.pdfESC Beyond Borders _From EU to You_ InfoPack general.pdf
ESC Beyond Borders _From EU to You_ InfoPack general.pdf
Fundacja Rozwoju Społeczeństwa Przedsiębiorczego
 
Supporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptxSupporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptx
Jisc
 
Operation Blue Star - Saka Neela Tara
Operation Blue Star   -  Saka Neela TaraOperation Blue Star   -  Saka Neela Tara
Operation Blue Star - Saka Neela Tara
Balvir Singh
 
The Roman Empire A Historical Colossus.pdf
The Roman Empire A Historical Colossus.pdfThe Roman Empire A Historical Colossus.pdf
The Roman Empire A Historical Colossus.pdf
kaushalkr1407
 

Recently uploaded (20)

Cambridge International AS A Level Biology Coursebook - EBook (MaryFosbery J...
Cambridge International AS  A Level Biology Coursebook - EBook (MaryFosbery J...Cambridge International AS  A Level Biology Coursebook - EBook (MaryFosbery J...
Cambridge International AS A Level Biology Coursebook - EBook (MaryFosbery J...
 
Synthetic Fiber Construction in lab .pptx
Synthetic Fiber Construction in lab .pptxSynthetic Fiber Construction in lab .pptx
Synthetic Fiber Construction in lab .pptx
 
Palestine last event orientationfvgnh .pptx
Palestine last event orientationfvgnh .pptxPalestine last event orientationfvgnh .pptx
Palestine last event orientationfvgnh .pptx
 
The Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official PublicationThe Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official Publication
 
Chapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptxChapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptx
 
Additional Benefits for Employee Website.pdf
Additional Benefits for Employee Website.pdfAdditional Benefits for Employee Website.pdf
Additional Benefits for Employee Website.pdf
 
TESDA TM1 REVIEWER FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
TESDA TM1 REVIEWER  FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...TESDA TM1 REVIEWER  FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
TESDA TM1 REVIEWER FOR NATIONAL ASSESSMENT WRITTEN AND ORAL QUESTIONS WITH A...
 
Instructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptxInstructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptx
 
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 
Digital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and ResearchDigital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and Research
 
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 
The Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve ThomasonThe Art Pastor's Guide to Sabbath | Steve Thomason
The Art Pastor's Guide to Sabbath | Steve Thomason
 
Home assignment II on Spectroscopy 2024 Answers.pdf
Home assignment II on Spectroscopy 2024 Answers.pdfHome assignment II on Spectroscopy 2024 Answers.pdf
Home assignment II on Spectroscopy 2024 Answers.pdf
 
Fish and Chips - have they had their chips
Fish and Chips - have they had their chipsFish and Chips - have they had their chips
Fish and Chips - have they had their chips
 
Unit 2- Research Aptitude (UGC NET Paper I).pdf
Unit 2- Research Aptitude (UGC NET Paper I).pdfUnit 2- Research Aptitude (UGC NET Paper I).pdf
Unit 2- Research Aptitude (UGC NET Paper I).pdf
 
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCECLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
CLASS 11 CBSE B.St Project AIDS TO TRADE - INSURANCE
 
ESC Beyond Borders _From EU to You_ InfoPack general.pdf
ESC Beyond Borders _From EU to You_ InfoPack general.pdfESC Beyond Borders _From EU to You_ InfoPack general.pdf
ESC Beyond Borders _From EU to You_ InfoPack general.pdf
 
Supporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptxSupporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptx
 
Operation Blue Star - Saka Neela Tara
Operation Blue Star   -  Saka Neela TaraOperation Blue Star   -  Saka Neela Tara
Operation Blue Star - Saka Neela Tara
 
The Roman Empire A Historical Colossus.pdf
The Roman Empire A Historical Colossus.pdfThe Roman Empire A Historical Colossus.pdf
The Roman Empire A Historical Colossus.pdf
 

CNN and its applications by ketaki

  • 1. P R E S E N T E D B Y PAT WA R I K E TA K I J . 2 0 1 7 M N S 0 1 0 U N D E R T H E G U I D A N C E O F D R . M R S . WA G H M A R E J . M .
  • 2. Contents  Introduction  ConvNet Layers  Activation functions  Applications of CNN  Image classification  Traffic sign recognition  Vehicle detection  Some another interesting application  Conclusion  References
  • 3. Introduction  What is ANN?  A neural network is a system of interconnected artificial “neurons” that exchange messages between each other.
  • 4. fig2. Training of neural network Fig3. Illustration of neuron and mathematical model
  • 5.  What is Convolutional neural network?  A CNN is a special case of the ANN . A CNN consists of one or more convolutional layers, often with a sub-sampling layer, which are followed by one or more fully connected layers as in a standard neural network.  convolution layers play the role of feature extractor Fig 4 Typical block diagram of a CNN
  • 6. ConvNet Layers 1) Convolution layers  The convolution operation extracts different features of the input. The first convolution layer extracts low-level features like edges, lines, and corners.
  • 7.
  • 9.  3) Non-linear layers  Neural networks in general and CNNs in particular rely on a non-linear “trigger” function to signal distinct identification of likely features on each hidden layer. CNNs may use a variety of specific functions such as rectified linear units (ReLUs) and continuous trigger (non-linear) functions to efficiently implement this non-linear triggering.  ReLU(Rectified Linear Unit) implements function y = max(x,0).
  • 10.  Continuous trigger (non-linear) function  The non-linear layer operates element by element in each feature. A continuous trigger function can be hyperbolic tangent (Figure I), absolute of hyperbolic tangent (Figure II), or sigmoid (Figure III). Figure I Figure II Figure III Figure IV : tanh processing
  • 12. Activation Functions  Activation functions are non-linarites that take on a single number and do some mathematical operations on it. 1. Sigmoid:  This non-linearity takes as input a real-valued function and outputs value in the range of 0 and 1. 2.Tanh:  Tanh can be considered as a scaled up version of sigmoid, outputting values in the range of -1 and 1. tanh (x) = 2f (2 x)−1 3. Leaky ReLU:  Slight modification to ReLU resulted in Leaky ReLU given by Equation.
  • 13. Applications of CNN  The main categories are as follows: 1. image and pattern recognition 2. speech recognition 3. natural language processing 4. video analysis
  • 15.  Adaboost can effectively control the quantity of weak classifiers, which greatly increases execution efficiency through integration; therefore, it has also been applied to real-time object testing.  This effectively reduces parameter redundancy, which results in greater generalization and higher training efficiency.  The first step in this paper is to introduce the basic knowledge of the convolution, pooling, weight decay, and dropout of the convolution neural network, as well as the theories and knowledge of ensemble learning.  With the convolution neural network and the learning principles of Adaboost, this paper takes the initiative to propose the Boost Convolutional Neural Network, and conducts experimentation and analysis of image classification with the benchmark datasets of CIFAR-10 image classification.
  • 16.  BOOSTING ENSEMBLE LEARNING  As shown in Figure below , ensemble learning completes learning tasks by constructing and cooperating with several learners, meaning the so-called ‘‘collectivism’’, where a group of individual learners are trained, and a strategy is then adopted to combine them.  Boosting Algorithm  Boosting is an algorithm concept that combines weak classifiers through certain methods to form a strong classifier with high classification performance.  Process of boosting classification: 1. Train the first weak classifier (h1) in the initialized training dataset 2. Combine the poorly-trained data of the weak classifier (h1) with new data to form the training data for a new round, and train the second weak classifier (h2)
  • 17. 3. Combine the poorly-trained data of weak classifier (h1) and weak classifier (h2) with new data to form the training data for a new round, and train the third weak classifier (h3) 4. Repeat Steps 2-3 until an adequate number of weak classifiers have been attained. 5. Form a strong classifier through the weighted voting of all weak classifiers.  ADABOOST ALGORITHM(Adaptive Boosting) As its name suggests, according to the trained classifiers, AdaBoost can self-adjust weak classifiers after learning and is sensitive to noise data and outliers. In some tasks, it can efficiently resist overfitting. The concept of Adaboost can be described, as follows:  the given training dataset T = {(x1, y1), (x2, y2) . . . (xN, yN)}, where x ∈ X, X ∈ R n , and yi is a mark set {−1, +1}.  Adaboost aims to acquire a series of weak classifiers from the training data, and then, combines these weak classifiers into a strong classifier. Moreover, its computational process can be described, as follows: 1. The weight distribution of the initial training data; each training sample is given the same initial weight: 1/N i.e
  • 18. D1 = (w11,w12 . . . w1i . . . ,w1N ), w1i = 1/ N i = 1, 2, . . . ,N 2. In the case of m = 1, 2, . . . , M uses the training data with the weight distribution (Dm) in learning to obtain a basic binomial classifier, as follows: Gm(x) : x → {−1, +1} 3. the rate of a wrong classification of classifier (Gm(x)) in the training dataset is, as follows: 4. coefficient (αm) of Gm(x)) is calculated, which refers to the voting weight of Gm(x)) in the final classier: 5. the weight distribution of the updated training datasets is, as follows:
  • 19. Zm is a normalized factor 6. the linear combination of basic classifiers is established: The final classifier is then obtained is:
  • 21.
  • 22.  SIMULATION EXPERIMENT AND ANALYSIS  To analyze the performance of the BoostCNN model, this chapter uses the four- layered convolution network and Softmax classifier as the feature learning architecture, and adopts the AdaBoost algorithm to generate several Softmax classifiers as the output.  CIFAR-10 DATASET  the CIFAR-10 dataset is an image dataset whose 60,000 colorful images (32 ∗ 32) are classified into 10 categories. Specifically, 50,000 images are classified as training data, while 10,000 images are classified as testing data. As shown in Figure 17, the 10 categories are plane, car, bird, cat, deer, dog, frog, horse, ship, and truck, and each category includes 6,000 images. Experimental comparison of CIFAR-10 testing datasets.
  • 23. CONCLUSION  This study focuses on how to combine a convolution neural network with AdaBoost to enhance the image identification performance of the learning algorithms.  After the convolution neural network is trained into the deep feature extraction model, and the original images are converted to acquire the deep features, AdaBoost is used for ensemble learning.  The conclusions of this paper are, as follows: 1) After the feature extraction of the deep convolution neural network, the original image data are fully abstracted, thus, the traditional learning algorithms can also effectively learn highly complicated image data. 2) Through the comparative experiments of the CIFAR-10 image dataset, AdaBoost can generally enhance the performance of base learners by 3%.
  • 25.  Traffic sign recognition (TSR) represents an important feature of advanced driver assistance systems, contributing to the safety of the drivers and vehicles as well.  Developing TSR systems requires the use of computer vision techniques, which could be considered fundamental in the field of pattern recognition in general. Despite all the previous works and research that has been achieved, traffic sign detection and recognition still remain a very challenging problem, precisely if we want to provide a real time processing solution.  In this paper, author present a comparative and analytical study of the two major approaches for traffic sign detection and recognition.  The first approach is based on the color segmentation technique and convolutional neural networks (C-CNN),  while the second one is based on the fast region-based convolutional neural networks approach (Fast R-CNN).
  • 26.  COLOR SEGMENTATION & CNN BASED APPROACH  The C-CNN method consists of selecting a set of regions for interest (ROIs) by applying a color tresholding on the input image, thus reducing the search space.  Then, a trained CNN is used to classify the ROI (whether it contains a traffic sign or not), followed by another CNN with the same architecture that is used to recognize the detected traffic signs.  Using the HSV features and grayscale color spaces, color thresholding can be applied to a given image or frame.  After color detection, an additional step is required in order to verify if the corresponding regions ROIs are a traffic sign or not, by using a CNN model as a classifier.
  • 27.
  • 28.
  • 29.  After converting RGB color space into HSV color space the CNN is applied as a model.  It is fairly simple and has 7 layers: 3 convolutional layers, 3 maxpooling layers for feature extraction and one fully connected layer as a classifier. The CNN was trained to recognize two classes: traffic sign / no traffic sign. For that, we merged two datasets: the GTSRB dataset and 30,000 random samples of the Cifar-10 dataset  After classifying the extracted ROIs, we apply a second CNN with same architecture of the first one, to recognize the detected traffic signs in the ROIs.  FAST R-CNN BASED APPROACH  Fast R-CNN was proposed to fix the disadvantages of R-CNN [28] and SPPnet, while improving their speed and accuracy. The Fast R-CNN method has several advantages: 1. Higher detection quality than R-CNN, and SPPnet, 2. Training is single-stage, using a multi-task loss. 3. No disk storage is required for feature caching. 4. Training can update all network layers.
  • 30.  The Fast R-CNN network takes as input an entire image and a set of object proposals that are calculated by an external algorithm. The network first processes the whole image with several convolutional (conv) and max pooling layers to produce a conv feature map. Then, for each object proposal a region of interest (ROI) pooling layer extracts a fixed-length feature vector from the feature map.  Each feature vector is fed into a sequence of fully connected(fc) layers that finally branch into two sibling output layers:  One that produces softmax probability estimates over K object classes plus a catch-all “background” class.  Another layer that outputs four real-value numbers for each of the K object classes. Each set of 4 values encodes refined bounding-box positions for one of the K classes.
  • 31.  EXPERIMENTAL RESULTS  In this case they used python language for implementation. For color segmentation, we used the OpenCv API and the well-known deep learning API Keras with Tensorflow backend were used to train the CNNs models in both approaches. 1. C-CNN APPROACH RESULTS Here they trained two CNNs with the same architecture, the first one for classifying the ROIs extracted after applying the color segmentation on input frames and the second one to recognize the detected traffic signs. The training time took 6 hours on an Intel core i7 3612QM 8GB 1T.
  • 32. 2. C-CNN APPROACH RESULTS  With the Fast R-CNN approach, we could achieve 94.8% accuracy on test set (the GTSDB). The training time took 3 days, on a server with a GTX 1080 TI ROC - 11GO graphic card  Even though the Fast R-CNN is a very good approach and showed interesting results in pattern detection & recognition in general, but if the training dataset is not balanced and big, that limits the performance of the approach.
  • 33.
  • 34.  CONCLUSION  This paper presented an analytical study of two effective and efficient road sign detection and recognition approaches.  The experimental results achieved after testing both of the methods on the German Traffic Sign Detection & Recognition datasets, conclude that the Fast R-CNN is so much faster than the C-CNN method, also it is invariant to illumination changes.  On the other hand, even though the C-CNN approach is slow and sensitive to weather conditions, it is invariant to scale and viewing angle.
  • 36.  Vehicle detection and counting in aerial images is important for a wide range of applications such as urban planning and traffic management.  Many methods have been introduced in the literature for solving this problem. These methods are either based on shallow learning or deep learning approaches. However, these methods suffer from relatively low precision and recall rate.  This paper introduces an automated vehicle detection and counting system in aerial images.  The proposed system utilizes convolution neural network (CNN) to regress a vehicle spatial density map across the aerial image. It has been evaluated on two publicly available datasets namely Munich and OIRDS.  The experimental results show that our proposed system is efficient and effective, and produces higher precision and recall rate than the comparative methods.
  • 37.
  • 38.  SHALLOW-LEARNING-BASED METHODS  The general strategy followed in this group relies on handcrafted features extraction followed by a classifier or cascade of classifiers.  authors proposed a system for car counting in aerial images captured by UAV. They have reduced search space by selecting the regions where cars might exist using a supervised classifier then extracted feature points using scale invariant feature transform (SIFT).  Then support vector machine (SVM) has been used in order to discriminate between the cars and all other objects. Four steps for car detection system have been introduced. 1. The proposed system starts with selecting the areas that might have cars. 2. Then, two sets of histogram of oriented gradients (HOG) features are extracted for vertical and horizontal filtering directions. 3. The discrimination between the cars and other objects has been performed by one of three suggested techniques: mutual information measure, normalized cross correlation, and combination of the correlation measure with SVM classification. 4. The discrimination is obtained by associating an orientation value to the points classified as cars. Finally, the points that belong to the same car are merged.
  • 39.  DEEP-LEARNING-BASED METHODS  Most of the works proposed in this category use convolution neural network for automatic features extraction. In deep convolutional neural network with multi-scale spatial pyramid pooling (SPP) has been employed in extracting the target patterns with different sizes. However, input images have been pre-processed by maximum normed gradient algorithm in order to restore the edges of the objects.  Another deep learning approach has been introduced. In this , the input image has been segmented into small homogeneous regions. Then the features in the segmented regions are extracted using pre-trained convolutional neural network (CNN) by a sliding-window approach.  Windows are classified using support vector machine (SVM) into car and no-car classes. Finally, post-processing is done such as morphological dilation to smooth the detected regions and fill the holes.
  • 40.  The proposed system 1. FULLY CONVOLUTIONAL REGRESSION NETWORK(FCRN) The proposed system. (a) Training phase. (b) Inference Phase.
  • 41.
  • 42.
  • 43. 3. IMPLEMENTATION DETAILS  The implementation of the proposed architecture is based on Tensorflow  During training phase, 224x224 random patches were selected from the aerial image. The selected patch contains at least one vehicle. Thus, patches with no vehicles were not chosen during training. In order to increase the amount of training examples, data augmentation techniques were utilized such as rotation, horizontal and vertical flipping and shifting.  The mean square error target function is used, as follows: where, X is the input patch with M samples, are all trainable parameters, YP is the predicted density map, and YT YT is the ground truth annotation.  RMSprop optimizer has been used for updating the parameter values.
  • 44.  DATASETS DESCRIPTION  The proposed system has been evaluated on two public datasets namely DLR Munich vehicle dataset provided by Remote Sensing Technology Institute of the German Aerospace Center and Overhead Imagery Research Data Set (OIRDS) dataset.  Munich dataset contains 20 images (5616 x 3744 pixels) taken by DLR 3K camera system at a height of 1000 m above the ground over the area of Munich, Germany. This dataset contains 3418 cars and 54 trucks annotated in the training image set and 5799 cars and 93 trucks annotated in testing image set.  OIRDS dataset This dataset contains 907 aerial images with approximately1800 annotated vehicles.
  • 45.  QUANTITATIVE EVALUATION AND COMPARISON F1- F1 score Performance Comparison between the proposed method and the state-of-the-art methods.
  • 46.  CONCLUSION  A novel vehicle detection and counting method has been introduced using convloutional regression neural network. In the proposed system, we have used regression model in order to predict the density map of the input patches. Then, the output of FCRN goes under empirical threshold which results a binary image.  Finally, a simple connected component algorithm is used for finding the locations and count of the blobs that represent the detected vehicles. The results of the proposed architecture outperforms the state-of-the-art methods.  have achieved the highest true positive rate and the lowest false alarm rate.
  • 47. Some another Interesting applications  Mixtures of Lightweight Deep Convolutional Neural Networks: Applied to Agricultural Robotics  Aircraft Type Recognition Based on Segmentation With Deep Convolutional Neural Networks  Deep Convolutional Neural Networks and Data Augmentation for Environmental Sound Classification  Fast Deep Neural Networks with Knowledge Guided Training and Predicted Regions of Interests for Real-time Video Object Detection
  • 48. Conclusion  CNNs give the best performance in pattern/image recognition problems and even outperform humans in certain cases. Cadence has achieved best-in-industry results using proprietary algorithms and architectures with CNNs.  Now we have studied CNN and its components along with the applications. Also we have studied three IEEE papers thoroughly that are having different applications that are Image classification, traffic sign recognition and Vehicle detection.  In previous section I have given more interesting applications you may go through it.
  • 49. References [1]Samer Hijazi, Rishi Kumar, and Chris Rowen, IP Group, Cadence,”Using Convolutional Neural Networks for Image Recognition”. [2]Neena Aloysius and Geetha M,” A Review on Deep Convolutional Neural Networks”, International Conference on Communication and Signal Processing, April 6-8, 2017, India. [3]Keiron O’Shea1 and Ryan Nash2,” An Introduction to Convolutional Neural Networks” [4] Kaoutar Sefrioui Boujemaa, Afaf Bouhoute, Karim Boubouh and Ismail Berrada,” Traffic sign recognition using convolutional neural networks”, 2017 International Conference on Wireless Networks and Mobile Communications (WINCOM). [5] SHIN-JYE LEE, TONGLIN CHEN, LUN YU, AND CHIN-HUI LAI,” Image Classification Based on the Boost Convolutional Neural Network”, Digital Object Identifier 10.1109/ACCESS.2018.2796722. [6] HILAL TAYARA, KIM GIL SOO , AND KIL TO CHONG ,“Vehicle Detection and Counting in High-Resolution Aerial Images Using Convolutional Regression Neural Network ”, Digital Object Identifier 10.1109/ACCESS.2017.DOI.