1. The document describes using a deep neural network to detect changes between two SAR images by preclassifying the images, training the neural network on selected samples, and analyzing the results.
2. A similarity matrix and variance matrix are calculated during preclassification to identify and jointly label similar pixels, while different pixels are labeled separately. Good samples are selected to train the neural network.
3. The neural network is tested on images with different types and levels of noise and performs well at change detection, with performance increasing as noise decreases. Future work could focus on accelerating the training process.
Pixel Recursive Super Resolution.
Ryan Dahl, Mohammad Norouzi & Jonathon Shlens
Google Brain.
Abstract
We present a pixel recursive super resolution model that
synthesizes realistic details into images while enhancing
their resolution. A low resolution image may correspond
to multiple plausible high resolution images, thus modeling
the super resolution process with a pixel independent conditional
model often results in averaging different details–
hence blurry edges. By contrast, our model is able to represent
a multimodal conditional distribution by properly modeling
the statistical dependencies among the high resolution
image pixels, conditioned on a low resolution input. We
employ a PixelCNN architecture to define a strong prior
over natural images and jointly optimize this prior with a
deep conditioning convolutional network. Human evaluations
indicate that samples from our proposed model look
Optimized Neural Network for Classification of Multispectral ImagesIDES Editor
The proposed work involves the multiobjective PSO
based optimization of artificial neural network structure for
the classification of multispectral satellite images. The neural
network is used to classify each image pixel in various land
cove types like vegetations, waterways, man-made structures
and road network. It is per pixel supervised classification using
spectral bands (original feature space). Use of neural network
for classification requires selection of most discriminative
spectral bands and determination of optimal number of nodes
in hidden layer. We propose new methodology based on
multiobjective particle swarm optimization (MOPSO) to
determine discriminative spectral bands and the number of
hidden layer node simultaneously. The result obtained using
such optimized neural network is compared with that of
traditional classifiers like MLC and Euclidean classifier. The
performance of all classifiers is evaluated quantitatively using
Xie-Beni and â indexes. The result shows the superiority of
the proposed method.
Pixel Recursive Super Resolution.
Ryan Dahl, Mohammad Norouzi & Jonathon Shlens
Google Brain.
Abstract
We present a pixel recursive super resolution model that
synthesizes realistic details into images while enhancing
their resolution. A low resolution image may correspond
to multiple plausible high resolution images, thus modeling
the super resolution process with a pixel independent conditional
model often results in averaging different details–
hence blurry edges. By contrast, our model is able to represent
a multimodal conditional distribution by properly modeling
the statistical dependencies among the high resolution
image pixels, conditioned on a low resolution input. We
employ a PixelCNN architecture to define a strong prior
over natural images and jointly optimize this prior with a
deep conditioning convolutional network. Human evaluations
indicate that samples from our proposed model look
Optimized Neural Network for Classification of Multispectral ImagesIDES Editor
The proposed work involves the multiobjective PSO
based optimization of artificial neural network structure for
the classification of multispectral satellite images. The neural
network is used to classify each image pixel in various land
cove types like vegetations, waterways, man-made structures
and road network. It is per pixel supervised classification using
spectral bands (original feature space). Use of neural network
for classification requires selection of most discriminative
spectral bands and determination of optimal number of nodes
in hidden layer. We propose new methodology based on
multiobjective particle swarm optimization (MOPSO) to
determine discriminative spectral bands and the number of
hidden layer node simultaneously. The result obtained using
such optimized neural network is compared with that of
traditional classifiers like MLC and Euclidean classifier. The
performance of all classifiers is evaluated quantitatively using
Xie-Beni and â indexes. The result shows the superiority of
the proposed method.
Fast Full Search for Block Matching Algorithmsijsrd.com
This project introduces configurable motion estimation architecture for a wide range of fast block-matching algorithms (BMAs). Contemporary motion estimation architectures are either too rigid for multiple BMAs or the flexibility in them is implemented at the cost of reduced performance. In block-based motion estimation, a block-matching algorithm (BMA) searches for the best matching block for the current macro block from the reference frame. During the searching procedure, the checking point yielding the minimum block distortion (MBD) determines the displacement of the best matching block.
Deep learning for image super resolutionPrudhvi Raj
Using Deep Convolutional Networks, the machine can learn end-to-end mapping between the low/high-resolution images. Unlike traditional methods, this method jointly optimizes all the layers of the image. A light-weight CNN structure is used, which is simple to implement and provides formidable trade-off from the existential methods.
Neural network based image compression with lifting scheme and rlceSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology.
PR-207: YOLOv3: An Incremental ImprovementJinwon Lee
TensorFlow Korea 논문읽기모임 PR12 207번째 논문 review입니다
이번 논문은 YOLO v3입니다.
매우 유명한 논문이라서 크게 부연설명이 필요없을 것 같은데요, Object Detection algorithm들 중에 YOLO는 굉장히 특색있는 one-stage algorithm입니다. 이 논문에서는 YOLO v2(YOLO9000) 이후에 성능 향상을 위하여 어떤 것들을 적용하였는지 하나씩 설명해주고 있습니다. 또한 MS COCO의 metric인 average mAP에 대해서 비판하면서 mAP를 평가하는 방법에 대해서도 얘기를 하고 있는데요, 자세한 내용은 영상을 참고해주세요~
논문링크: https://arxiv.org/abs/1804.02767
영상링크: https://youtu.be/HMgcvgRrDcA
This is a presentation on Handwritten Digit Recognition using Convolutional Neural Networks. Convolutional Neural Networks give better results as compared to conventional Artificial Neural Networks.
Image classification is perhaps the most important part of digital image analysis. In this paper, we compare the most widely used model CNN Convolutional Neural Network , and MLP Multilayer Perceptron . We aim to show how both models differ and how both models approach towards the final goal, which is image classification. Souvik Banerjee | Dr. A Rengarajan "Hand-Written Digit Classification" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd42444.pdf Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42444/handwritten-digit-classification/souvik-banerjee
Large Convolutional Network models have
recently demonstrated impressive classification
performance on the ImageNet benchmark
(Krizhevsky et al., 2012). However
there is no clear understanding of why they
perform so well, or how they might be improved.
In this paper we address both issues.
We introduce a novel visualization technique
that gives insight into the function of intermediate
feature layers and the operation of
the classifier. Used in a diagnostic role, these
visualizations allow us to find model architectures
that outperform Krizhevsky et al. on
the ImageNet classification benchmark. We
also perform an ablation study to discover
the performance contribution from different
model layers. We show our ImageNet model
generalizes well to other datasets: when the
softmax classifier is retrained, it convincingly
beats the current state-of-the-art results on
Caltech-101 and Caltech-256 datasets
Offline Character Recognition Using Monte Carlo Method and Neural Networkijaia
Human Machine interface are constantly gaining improvements because of increasing development of
computer tools. Handwritten Character Recognition do have various significant applications like form
scanning, verification, validation, or checks reading. Because of the importance of these applications
passionate research in the field of Off-Line handwritten character recognition is going on. The challenge in
recognising the handwritings lies in the nature of humans, having unique styles in terms of font, contours,
etc. This paper presents a novice approach to identify the offline characters; we call it as character divider
approach which can be used after pre-processing stage. We devise an innovative approach for feature
extraction known as vector contour. We also discuss the pros and cons including limitations, of our
approach
Road Network Extraction using Satellite Imagery.SUMITRAJ312049
This is my Internship project ppt on Road Network Extraction Using Satellite Imagery.
In this project, A robust and efficient method for the extraction of roads from a given set of satellite images is explained.
In this work, we implement the U-Net segmentation architecture on the Mnih et. al.Massachusetts Roads Dataset for the task of road network extraction.
Fast Full Search for Block Matching Algorithmsijsrd.com
This project introduces configurable motion estimation architecture for a wide range of fast block-matching algorithms (BMAs). Contemporary motion estimation architectures are either too rigid for multiple BMAs or the flexibility in them is implemented at the cost of reduced performance. In block-based motion estimation, a block-matching algorithm (BMA) searches for the best matching block for the current macro block from the reference frame. During the searching procedure, the checking point yielding the minimum block distortion (MBD) determines the displacement of the best matching block.
Deep learning for image super resolutionPrudhvi Raj
Using Deep Convolutional Networks, the machine can learn end-to-end mapping between the low/high-resolution images. Unlike traditional methods, this method jointly optimizes all the layers of the image. A light-weight CNN structure is used, which is simple to implement and provides formidable trade-off from the existential methods.
Neural network based image compression with lifting scheme and rlceSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology.
PR-207: YOLOv3: An Incremental ImprovementJinwon Lee
TensorFlow Korea 논문읽기모임 PR12 207번째 논문 review입니다
이번 논문은 YOLO v3입니다.
매우 유명한 논문이라서 크게 부연설명이 필요없을 것 같은데요, Object Detection algorithm들 중에 YOLO는 굉장히 특색있는 one-stage algorithm입니다. 이 논문에서는 YOLO v2(YOLO9000) 이후에 성능 향상을 위하여 어떤 것들을 적용하였는지 하나씩 설명해주고 있습니다. 또한 MS COCO의 metric인 average mAP에 대해서 비판하면서 mAP를 평가하는 방법에 대해서도 얘기를 하고 있는데요, 자세한 내용은 영상을 참고해주세요~
논문링크: https://arxiv.org/abs/1804.02767
영상링크: https://youtu.be/HMgcvgRrDcA
This is a presentation on Handwritten Digit Recognition using Convolutional Neural Networks. Convolutional Neural Networks give better results as compared to conventional Artificial Neural Networks.
Image classification is perhaps the most important part of digital image analysis. In this paper, we compare the most widely used model CNN Convolutional Neural Network , and MLP Multilayer Perceptron . We aim to show how both models differ and how both models approach towards the final goal, which is image classification. Souvik Banerjee | Dr. A Rengarajan "Hand-Written Digit Classification" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd42444.pdf Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42444/handwritten-digit-classification/souvik-banerjee
Large Convolutional Network models have
recently demonstrated impressive classification
performance on the ImageNet benchmark
(Krizhevsky et al., 2012). However
there is no clear understanding of why they
perform so well, or how they might be improved.
In this paper we address both issues.
We introduce a novel visualization technique
that gives insight into the function of intermediate
feature layers and the operation of
the classifier. Used in a diagnostic role, these
visualizations allow us to find model architectures
that outperform Krizhevsky et al. on
the ImageNet classification benchmark. We
also perform an ablation study to discover
the performance contribution from different
model layers. We show our ImageNet model
generalizes well to other datasets: when the
softmax classifier is retrained, it convincingly
beats the current state-of-the-art results on
Caltech-101 and Caltech-256 datasets
Offline Character Recognition Using Monte Carlo Method and Neural Networkijaia
Human Machine interface are constantly gaining improvements because of increasing development of
computer tools. Handwritten Character Recognition do have various significant applications like form
scanning, verification, validation, or checks reading. Because of the importance of these applications
passionate research in the field of Off-Line handwritten character recognition is going on. The challenge in
recognising the handwritings lies in the nature of humans, having unique styles in terms of font, contours,
etc. This paper presents a novice approach to identify the offline characters; we call it as character divider
approach which can be used after pre-processing stage. We devise an innovative approach for feature
extraction known as vector contour. We also discuss the pros and cons including limitations, of our
approach
Road Network Extraction using Satellite Imagery.SUMITRAJ312049
This is my Internship project ppt on Road Network Extraction Using Satellite Imagery.
In this project, A robust and efficient method for the extraction of roads from a given set of satellite images is explained.
In this work, we implement the U-Net segmentation architecture on the Mnih et. al.Massachusetts Roads Dataset for the task of road network extraction.
CONVOLUTIONAL NEURAL NETWORK BASED RETINAL VESSEL SEGMENTATIONCSEIJJournal
In human eye, the state of the blood vessel is a crucial diagnostic factor. The segmentation of blood vessel
from the fundus image is difficult due to the spatial complexity, adjacency, overlapping and variability of
blood vessel. The detection of ophthalmic pathologies like hypertensive disorders, diabetic retinopathy and
cardiovascular diseases are remain challenging task due to the wide-ranging distribution of blood vessels.
In this paper, Stacked Autoencoder and CNN (Convolutional Neural Network) technique is proposed to
extract the blood vessel from the fundus image. Based on the experiments conducted using the Stacked
Autoencoder and Convolutional Neural Network gives 90% & 95% accuracy for segmentation.
Convolutional Neural Network based Retinal Vessel SegmentationCSEIJJournal
In human eye, the state of the blood vessel is a crucial diagnostic factor. The segmentation of blood vessel
from the fundus image is difficult due to the spatial complexity, adjacency, overlapping and variability of
blood vessel. The detection of ophthalmic pathologies like hypertensive disorders, diabetic retinopathy and
cardiovascular diseases are remain challenging task due to the wide-ranging distribution of blood vessels.
In this paper, Stacked Autoencoder and CNN (Convolutional Neural Network) technique is proposed to
extract the blood vessel from the fundus image. Based on the experiments conducted using the Stacked
Autoencoder and Convolutional Neural Network gives 90% & 95% accuracy for segmentation.
Image reconstruction through compressive sampling matching pursuit and curvel...IJECEIAES
An interesting area of research is image reconstruction, which uses algorithms and techniques to transform a degraded image into a good one. The quality of the reconstructed image plays a vital role in the field of image processing. Compressive Sampling is an innovative and rapidly growing method for reconstructing signals. It is extensively used in image reconstruction. The literature uses a variety of matching pursuits for image reconstruction. In this paper, we propose a modified method named compressive sampling matching pursuit (CoSaMP) for image reconstruction that promises to sample sparse signals from far fewer observations than the signal’s dimension. The main advantage of CoSaMP is that it has an excellent theoretical guarantee for convergence. The proposed technique combines CoSaMP with curvelet transform for better reconstruction of image. Experiments are carried out to evaluate the proposed technique on different test images. The results indicate that qualitative and quantitative performance is better compared to existing methods.
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
MEDIAN BASED PARALLEL STEERING KERNEL REGRESSION FOR IMAGE RECONSTRUCTIONcscpconf
Image reconstruction is a process of obtaining the original image from corrupted data. Applications of image reconstruction include Computer Tomography, radar imaging, weather
forecasting etc. Recently steering kernel regression method has been applied for image reconstruction [1]. There are two major drawbacks in this technique. Firstly, it is computationally intensive. Secondly, output of the algorithm suffers form spurious edges (especially in case of denoising). We propose a modified version of Steering Kernel Regression called as Median Based Parallel Steering Kernel Regression Technique. In the proposed algorithm the first problem is overcome by implementing it in on GPUs and multi-cores. The second problem is addressed by a gradient based suppression in which median filter is used. Our algorithm gives better output than that of the Steering Kernel Regression. The results are
compared using Root Mean Square Error(RMSE). Our algorithm has also shown a speedup of 21x using GPUs and shown speedup of 6x using multi-cores.
MEDIAN BASED PARALLEL STEERING KERNEL REGRESSION FOR IMAGE RECONSTRUCTIONcsandit
Image reconstruction is a process of obtaining the original image from corrupted data.Applications of image reconstruction include Computer Tomography, radar imaging, weather forecasting etc. Recently steering kernel regression method has been applied for image reconstruction [1]. There are two major drawbacks in this technique. Firstly, it is computationally intensive. Secondly, output of the algorithm suffers form spurious edges(especially in case of denoising). We propose a modified version of Steering Kernel Regression called as Median Based Parallel Steering Kernel Regression Technique. In the proposed algorithm the first problem is overcome by implementing it in on GPUs and multi-cores. The second problem is addressed by a gradient based suppression in which median filter is used.Our algorithm gives better output than that of the Steering Kernel Regression. The results are compared using Root Mean Square Error(RMSE). Our algorithm has also shown a speedup of 21x using GPUs and shown speedup of 6x using multi-cores.
Median based parallel steering kernel regression for image reconstructioncsandit
Image reconstruction is a process of obtaining the original image from corrupted data.
Applications of image reconstruction include Computer Tomography, radar imaging, weather
forecasting etc. Recently steering kernel regression method has been applied for image
reconstruction [1]. There are two major drawbacks in this technique. Firstly, it is
computationally intensive. Secondly, output of the algorithm suffers form spurious edges
(especially in case of denoising). We propose a modified version of Steering Kernel Regression
called as Median Based Parallel Steering Kernel Regression Technique. In the proposed
algorithm the first problem is overcome by implementing it in on GPUs and multi-cores. The
second problem is addressed by a gradient based suppression in which median filter is used.
Our algorithm gives better output than that of the Steering Kernel Regression. The results are
compared using Root Mean Square Error(RMSE). Our algorithm has also shown a speedup of
21x using GPUs and shown speedup of 6x using multi-cores.
Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using A...ijtsrd
This study proposes Artificial Neural Network ANN based field strength prediction models for the rural areas of Abuja, the federal capital territory of Nigeria. The ANN based models were created on bases of the Generalized Regression Neural network GRNN and the Multi Layer Perceptron Neural Network MLP NN . These networks were created, trained and tested for field strength prediction using received power data recorded at 900MHz from multiple Base Transceiver Stations BTSs distributed across the rural areas. Results indicate that the GRNN and MLP NN based models with Root Mean Squared Error RMSE values of 4.78dBm and 5.56dBm respectively, offer significant improvement over the empirical Hata Okumura counterpart, which overestimates the signal strength by an RMSE value of 20.17dBm. Deme C. Abraham ""Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using Artificial Neural Networks"" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-2 , February 2020,
URL: https://www.ijtsrd.com/papers/ijtsrd30228.pdf
Paper Url : https://www.ijtsrd.com/computer-science/artificial-intelligence/30228/mobile-network-coverage-determination-at-900mhz-for-abuja-rural-areas-using-artificial-neural-networks/deme-c-abraham
New Approach of Preprocessing For Numeral RecognitionIJERA Editor
The present paper proposes a new approach of preprocessing for handwritten, printed and isolated numeral
characters. The new approach reduces the size of the input image of each numeral by discarding the redundant
information. This method reduces also the number of features of the attribute vector provided by the extraction
features method. Numeral recognition is carried out in this work through k nearest neighbors and multilayer
perceptron techniques. The simulations have obtained a good rate of recognition in fewer running time.
Electroencephalography (EEG) based brain Computer interface (BCI) needs efficient algorithms to extract discriminative features from raw EEG signals. The issue of selecting optimizing spatial spectral features is key to high performance motor imagery(MI) classification, which is one of the main topics in EEG-based brain computer interfaces. Some novel methods are used first which formulates the selection of features as maximizing mutual information between class labels and features. It then uses an efficient algorithms for pattern feature extraction frame work,to select an effective feature set. The results shows the classification accuracy obtained and is compared with the other existing algorithms
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
1. 1. Calculate a similarity matrix of the original two images based
on:
𝑆𝑖𝑗 =
|𝐼𝑖𝑗
1
− 𝐼𝑖𝑗
2
|
𝐼𝑖𝑗
1
+ 𝐼𝑖𝑗
2
2. Calculate a variance matrix for each of the original two images
based on:
𝛿𝑖𝑗
2
= 𝐼𝑖𝑗
1
𝐼𝑖𝑗
1
𝐼𝑖𝑗
2
𝐼𝑖𝑗
1
+ 𝐼𝑖𝑗
2 [𝑆𝑖𝑗]2
1. Iterate over all 𝐼𝑖𝑗, if 𝑆𝑖𝑗 > 𝑇, where 𝑇 represents an iterative
threshold, then jointly label 𝐼𝑖𝑗
1
and 𝐼𝑖𝑗
2
by FCM based on the
principle of minimum variance. Otherwise label 𝐼𝑖𝑗
1
and 𝐼𝑖𝑗
2
separately.
2. Pick good samples to feed to the neural network. We pick the
“good” pixels based on a comparison between its label and the
labels of the pixels in the neighborhood around it.
𝑄(𝑝 𝜉𝑛 𝜖 𝑁𝑖𝑗 ⋀ Ω 𝜉𝑛 = Ω𝑖𝑗)
𝑛 × 𝑛
> 𝑎
𝐼𝑖𝑗
𝑛
represents the gray level of the nth image at
the position (i, j)
Introduction
Deep Neural Networks
Conclusion and Future Work
Image change detection involves identifying the changes that have
occurred between two images of a specific area over different time
periods. It is an important problem for both civil and military
applications. Synthetic Aperture Radar (SAR) images are
especially difficult to analyze, since these satellite images produce
an abundance of speckle noise. Current methods involve
generating a Difference Image (DI) and analyzing the DI. This
project attempts to apply the concept of neural networks to detect
changes between two images, avoiding the process of analyzing a
DI and/or proactively reducing noise. The process took form in 3
steps: preclassifying before and after SAR images to obtain good
samples to train the network with, creating and training networks,
and analyzing results of the network. Parts of the process are also
accelerated through principles of parallelization.
Preclassification Results
A DNN is a mathematical model to represent feature recognition.
The neural network consists of a network of nodes in layers,
where certain nodes are connected. These connections have
different weights and these nodes have biases. An activation of a
node can, in turn, activate a connected node based on the
following function:
𝜎( 𝑊𝑖 𝑣𝑖 + 𝑐𝑖)
The weights of the connections are initially set randomly. The input
layer of nodes are set as the features of the good sample
neighborhoods. After updating the states of all nodes in the
network, the neural network reconstructs a set of input states
based on the states of the output node. The weights are then
updated based on the following function:
𝜀( 𝑣𝑖ℎ𝑗 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 − 𝑣𝑖ℎ𝑗 𝑟𝑒𝑐𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑒𝑑)
We trained a restricted Boltzmann machine network (RBM), which
consists of a type of layer-by-layer training that restricts nodes
from communicating in their own layer.
Supported by NSF Grant to the REU EXERCISE - Explore
Emerging Computing in Science and Engineering Program
Compared to traditional image segmentation methods, the neural
network performed quite well. The network’s performance
increases when the amount of noise decreases. We did not
discover consistent performance with respect to different types of
noise. Based on the results, the network can interpret images with
Gaussian noise, speckle noise, Poisson noise, and salt and
pepper noise.
A few image segmentation techniques were tested for
preclassification. FCM was concluded to be the most accurate.
However, clustering and thresholding fails to take into account
spatial features on an image. Other segmentation techniques that
do account for spatial features include edge detection and region
growth. Some of these require long processing times and/or
human intervention, but can be tested in the future.
Another area of future work lies in accelerating the training of the
neural network. Currently, parallelization affects only iterative
image processing. Future work could be put towards discovering a
parallel structure for the network training.
Neural Network Results
M. Gong, J. Zhao, J. Liu, Q. Miao and L. Jiao, "Change detection in
synthetic aperture radar images based on deep neural networks", IEEE
Trans. Neural Netw. Learn. Syst.
Masayuki Tanaka and Masatoshi Okutomi, A Novel Inference of a
Restricted Boltzmann Machine, International Conference on Pattern
Recognition (ICPR2014), August, 2014.
P. L. Rosin and E. Ioannidis, "Evaluation of global image thresholding for
change detection", Pattern Recognit. Lett., vol. 24, no. 14, pp. 2345-
2356, 2003
J. C. Dunn (1973): "A Fuzzy Relative of the ISODATA Process and Its
Use in Detecting Compact Well-Separated Clusters", Journal of
Cybernetics 3: 32-57
FCM Clustering
1 2
3 4
1: Before: Santarem, Brazil (June 2007)
2: After: Santarem, Brazil (June 2008)
3: Similarity Matrix [Darker -> Less
Similar]
4: Result of thresholding [Dark; Expected
Unchanged || Light; Expected Change]
5: Selected Good Samples [Represented
by light pixels]
𝜎 represents the logistic function,
1
1+𝑒−𝑥
𝑊𝑖 represents the weight of the connection
𝑣𝑖 represents the state of the input node
𝑐𝑖 represents the bias of the connection
Accelerated Change Detection in Synthetic Aperture Radar Images
based on Deep Neural Networks
Lizzie Koshelev*, Malcolm Milton*, Frank Liao*, Yuanwei Jin**, Enyue Lu†
*Department of Mathematics & Computer Science, Salisbury University, Salisbury, MD; **Department of Engineering and Aviation Sciences, University of
Maryland Eastern Shore, Princess Anne, MD; †Department of Mathematics & Computer Science, Salisbury University, Salisbury, MD
5
den. represents similar pixels
num. represents neighborhood size
α represents a chosen threshold
𝑣𝑖 represents input state
ℎ𝑗 represents output state
𝜀 represents a chosen learning rate
FCM is a popular image segmentation technique that segments an
image by discovering cluster centers.
Main objective of fuzzy c-means algorithm is to minimize:
𝐽 = (𝜇𝑖𝑗) 𝑚
𝑑𝑖𝑗
2
𝑐
𝑗=1
𝑛
𝑖=1
1) Randomly select 𝑐 cluster centers.
2) Calculate the fuzzy memberships 𝜇𝑖𝑗 using:
𝜇𝑖𝑗 =
1
(
𝑑𝑖𝑗
𝑑𝑖𝑘
)
2
𝑚−1𝑐
𝑘=1
3) Compute the fuzzy centers 𝑣𝑗 using:
𝑣𝑗 =
𝜇𝑖𝑗x𝑖
𝑛
𝑖=1
𝜇𝑖𝑗
𝑛
𝑖=1
4) Repeat steps 2) and 3) until the minimum 𝐽 is achieved or until
the update change of membership values is deemed negligible.
𝑥𝑖 is the ith data element
𝑣𝑗 is the jth center
𝑛 is the number of data points.
𝑚 is the fuzziness index, 𝑚 ∈ [1, ∞].
𝑐 is the number of cluster center.
𝜇𝑖𝑗 is the strength of 𝑥𝑖 belonging to 𝑣𝑗
𝑑𝑖𝑗 is the Euclidean distance between 𝑥𝑖 and 𝑣𝑗
2
3
4
Preclassification
PCC TP (%) TN (%) FP (%) FN (%)
Puppy (Artificial) 98.21 3.80 94.41 1.78 0.01
Santarem 96.00 1.15 94.86 3.39 0.61
River 97.11 1.10 96.00 2.25 0.64
Santarem 2 87.68 27.79 59.90 6.16 6.16
𝑃𝐶𝐶 =
𝑇𝑃 + 𝑇𝑁
𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁
TP – correctly classified as changed
TN – correctly classified as unchanged
FP – incorrectly classified as changed
FN – incorrectly classified as unchanged
Neural Network Results
Performance of Neural Network against Images
with Varying Levels of Artificial Noise:
Variance PCC Network Change Map
0 99.5
5 93.6
10 89.9
Before After
Speckle Variance 5 Ground Truth Change
Performance of Neural Network utilizing SAR
Data Sets and Artificial Images:
Neural Network Results
0
0.2
0.4
0.6
0.8
1
1.2
0 2 4 6 8 10
PCC
Set of Images
PCC of Different Noise Types
on Different Sets of Images
Speckle
Salt & Pepper
Gaussian
Poisson
Speckle Salt and Pepper
Gaussian Poisson
Performance of Neural Network against
Different Types of Noise:
Method:
1) Normalize the images using the Frobenious norm,
∥ 𝐴 ∥ 𝐹≡ |𝑎𝑖𝑗|2
𝑛
𝑗=1
𝑚
𝑖=1
2) Divide each image by its Frobenious norm and multiply that of the speckled
image.
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
0 2 4 6 8 10 12
PCC
Variance
PCC v. Variance of Speckle Noise
Set 1
Set 2
Set 3
Set 4
Set 5
Set 6
Set 7
Set 8
Set 9
1: Puppy Image w/ Artificial Noise and Change
2: Santarem, Brazil (June 2007 & May 2008)*
3: Yellow River Estuary (June 2008 & June 2009)
4: Santarem, Brazil (June 2007 & May 2008)*
*SAR Images courtesy of NASA Spatial Data Access Tool
References
Before Image After Image Neural Network
Change Map
Ground Truth
Change Map
1
Acknowledgements
Parallelization of Code:
We parallelized our code with Matlab’s Parallel Processing Toolbox to
speed the process of sorting data to train and test the network.
Time Before (hours) Time After (hours) Percent of Original Time
3.27 1.07 32.7