Change Detection For Remote Sensing Image Based On Spatial Neighborhood Information

Remote sensing image change detection is mainly to detect the change area of remote sensing image which is in the same area of different phases via distinct change detection methods and analyze the change detection results.Comparisons before classification is the mainstream research pattern of change detection,it mains that we need to firstly generate the difference image which contains the change information of two or multi-temporal remote sensing image,and then analyze the difference image for obtaining the change detection results.Due to the noise carried by remote sensing image when imaging will interfere the change information and affect the accuracy of detection results.Althougt the traditional MRF model considers the spatial neighborhood relation among pixels and reduces the interference of noise to change detection results,the spatial neighborhood relation among pixels is directly defined as 0 or 1 by Potts model in traditional MRF model,which easily causes overinterpretation of spatial neighborhood relation among pixels.Therefore,there are two key issues that need to be solved in this thesis: definng the saptial neighborhood relation reasonably and reducing the impact of noise on change detection results.In order to sovle these two problems,a new MRF model is cited in this thesis.This new MRF model introduces the spatial gravity model,the probabilistic constraints between pixels and classification category,and the spatial distance relation into Potts model to redefine the spatial neighborhood relation among pixels.The new MRF model still uses the clustering center which obtained by initial classification algorithm to define two thresholds,and the two thresholds will distinguish the difference image into three regions of unchanged,uncertain and changed.The penalty coefficient in Potts is also redefined by these thresholds.This thesis also uses two filter measures based on morphology and MRF to optimize the change detection results.By the analysis of comparative experiment of change detection of eight SAR image data sets and the change detection comparative experiments between two different difference images in the same data set,the effectiveness of the method described in this thesis is verified.