Change Detection For SAR Image Based On Neighborhood Relative Entropy And Fusing Texture Information

Problems such as the transformation and development of cities,changes in land use/coverage,and frequent occurrence of natural disasters will cause changes in the attributes and nature of land cover.It has a great significance to timely and effectively detect the change of surface features information such as location,change process,details.Synthetic Aperture Radar(SAR)has increasingly become an important data source for dynamic monitoring for remote sensing,because of its all-weather day and night imaging capabilities.SAR image change detection is the process of analyzing and extracting the changes by multitemporal images acquired in the same geographical area at different times.It has been a hot spot and an important research direction in the field of earth observation.However,it is faces several problems for change detection by using SAR image:(1)the influence of speckle noise;(2)the performance of difference image;(3)how to get the best threshold.To solve these questions,we construct more robust difference image and studied better preformance change detection algorithm by considering gray information,texture feature and spatial relativity.The main work and innovation are as follows:1.This paper proposes a method of SAR image change detection based on neighbor relative entropy to solve the problem of speckle noise in traditional algebraic operation method.This method utilizes the relative entropy operation and introduces the different forms of neighborhood information to fully combine the relationship between two-phase SAR image pixel and their neighborhood gray scales.The degree of similarity of the corresponding neighborhood block is used to characterize the degree of difference in two-phase SAR image.Then,the FLICM algorithm is used for classifying changed and unchanged areas and the change detecting map can be obtained automatically.Experiment results show that this method is robust to speckle noise.2.The change detection approach of feature level is studied to make best of texture information in high resolution SAR image.First,we obtain texture difference image through texture feature which calculated by gray level co-occurrence matrix.Then,the probability distribution of unchanged pixel and changed pixel in texture difference image were estimated by generalized Gaussian model.Final,unsupervised change detection on multi-temporal SAR images implements by combining KI threshold algorithm to obtain the best segmentation threshold.3.A change detection approach combined intensity-texture feature based on adaptive neighborhood mean filter is proposed without losing local information.First,we denoise intensity imagery by an improved adaptive neighborhood mean filtering method using twenty-one neighborhood forms.Then,we use a discrete stationary wavelet transform(SWT)as a tool to fuse the difference map of the log ratio and the I DM texture difference map,and then extracts the change region based on the Markov random field model segmentation method to obtain the final change detection binary result map.Through comparative analysis of other methods,it is proved that this method has good detection performance for new buildings.