SAR Images Change Detection Method Based On Low-Rank And Sparse Model

SAR has capabilities of high resolution and all-weather monitoring.Change detection for SAR images is a process that analyzes the changed areas between a pair of images acquired on the same area but different times,the main task is to determine whether the targets in the area have changed.In recent years,SAR images change detection techniques have been used in the landscape,natural disaster monitoring and evaluation as well as military monitoring.In this thesis,we address the issue for SAR images change detection based on the low-rank representation and low-rank and sparse matrix factorization method.Low-rank representation and low-rank and sparse matrix factorization theories are important techniques in the compression sensing,the application field is very broad,such as image processing,computer vision,data dimension reduction and so on.In this thesis,we apply low-rank representation and low-rank and sparse matrix factorization to SAR images change detection,and we propose two methods under the low-rank and sparse model for SAR images change detection.The major works can be summarized as follows:Analysis of the post-classification comparison and low-rank representation theories for SAR images change detection.The purpose of change detection is analyzing the information of pixels between the images,which is divided the difference information into the changed areas and the non-changed areas.This method assumes that there is a highly similar lowrank matrix between the two images,and the variation region data distribution is sparseness,and the non-changed region data distribution has low-rank quality.This method obtains the initial low-rank matrix and the sparse matrix through the post-classification comparison method,then calculates the sparse matrix which stand for changed areas by randomized lowrank and sparse matrix decomposition.The experimental results show that the method has a good change test result.Analysis of the low-rank and sparse matrix factorization and Bayes classifier,we propose a novel method for SAR images change detection based on low-rank and sparse information.Existed images change detection method did not make full use of the low-rank information in the difference image,therefore,this method uses the Bayes classifier to cope with that problem.In order to guarantee the classification performance of the change detection algorithm,this method obtains the low-rank and sparse matrix by low-rank and sparse matrix factorization,then it takes full use of low-rank and sparse matrix by Bayes classifier rather than focuses solely on the sparse information.In this method,it first trains the Bayes classifier by computing the expectation and variance of the low-rank and sparse matrix data sets,then analyzes the difference image to get the final result.Experiments show that this method can greatly reduce the number of false positives,and improve the overall accuracy of the change detection and classification accuracy.