Research On SAR Image Change Detection Method Based On Capsule Networ

The change detection of synthetic aperture radar(SAR)image aims to identify the changed regions by analyzing two images over the same area acquired at different times.SAR images are not affected by external conditions such as light,weather.It can achieve high-resolution,large-area detection of ground targets.Therefore,SAR images have a good development prospect.It is widely used in the early warning of natural disasters,agricultural production,urban planning,military deployment and other fields.However,the accuracy of change detection is negatively affected by the inherent speckle noise in SAR images.To solve this problem,a method based on sparse representation,saliency detection and a capsule network is proposed taking into account the characteristics of the SAR data sets.It improves the accuracy of change detection.The main contents in this thesis are summarized as below:1.In order to reduce the influence of speckle noise in the input images on the accuracy of change detection,a SAR image change detection method based on sparse representation and a capsule network is proposed.Firstly,sparse representation is used to extract the sparse features of the difference image.Secondly,the fuzzy C-means clustering method is used to classify the difference image.We select some highconfidence samples from the difference image as the training samples,and the others as the test samples.Finally,a lightweight capsule network is constructed.Fewer convolutional kernels are utilized in the lightweight capsule network,and the training parameters are significantly decreased.The network is used to mine the spatial relationship between features.The trained network can classify the changed and unchanged pixels.Experiments are performed on four SAR data sets to compare the proposed method with related methods.The results shows that the proposed method has good performance in suppressing speckle noise.It can improve the accuracy of SAR image change detection.2.To balance the number of changed samples and unchanged samples,a method of SAR image change detection based on saliency detection and an attention capsule network is proposed.Firstly,the method of saliency similarity detection is used to process the difference image.The significant changed areas on the difference image are extracted.Taking into account the characteristics of saliency detection,we only select training samples in the salient areas of the difference image.The pixels in the background areas are ignored.It greatly reduces the number of training samples.Secondly,we build a capsule network based on the attention mechanism.The spatial attention model is used to extract key features.The capsule network achieves accurate classification.Finally,the change map is generated combined with the classification results of the capsule network.Experimental results on several real SAR data sets confirm the proposed method can increase the accuracy for change detection.