Change Detection Of Polarimetric SAR Image Based On HLT Difference Map Analysis

The change detection of polarimetric SAR image refers to the process of analyzing and quantifying the changes of the same scene at two times.This process makes full use of the scattering characteristics of ground objects.It has a wide application prospect in land resources monitoring,forest vegetation cover and natural disaster analysis,and has important research value.In the change detection of polarimetric SAR image,the inherent speckle noise of polarimetric SAR image will reduce the quality of difference image and destroy the texture information.Therefore,the key problem of polarimetric SAR image change detection is how to maintain the texture information in the difference map to achieve accurate change location and effectively suppress speckle noise interference.In this paper,Hotelling Lawley trace(HLT)difference operator is used to extract the difference information in two time-phase polarimetric SAR images and generate HLT difference map.HLT difference map can completely describe the changes in two time-phase SAR images,and has rich local detail information,but the speckle noise level in the difference map is relatively high.This paper focuses on how to accurately extract the change region in the HLT difference map and effectively suppress the speckle noise interference,and carries out the following research:Chapter 3 analyzes the change information in the difference map from the perspective of multi-scale,and then proposes a polarimetric SAR change detection method based on difference map complementarity and point line singularity fusion.On the premise of obtaining the HLT difference map,this method further constructs the Shannon entropy difference map with low speckle noise level and complete overall structure contour information.Then,the multi-scale Stationary Wavelet Transform(SWT)with strong point singularity detection ability is used to decompose the HLT difference map,and the multiscale SWT low-frequency images are used to suppress the noise of the HLT difference map and extract the main change region;The Shannon entropy difference map is decomposed by the multi-scale and multi-directional Non-Subsampled Shearlet Transform(NSST),which has strong ability of line singularity detection,and the contour information such as curve edge is extracted to make up for the missing high-frequency information of the multi-scale SWT low-frequency images.Then,the extracted complementary information is fused in scale based on Fourth Order Partial Differential Equations(FPDEs)and Principal Component Analysis(PCA),and the fused image with low noise level and complete and accurate change information is obtained.Finally,the Bayesian threshold segmentation method is introduced to thresholding the fusion results within the scale,and the generated binary images are formed into the final change detection result by "And operation".Through the simulation experiment and analysis of four groups of measured data,the effectiveness of the algorithm in this chapter is proved.Chapter 4 combines the statistical model and random field theory,and then proposes a polarimetric SAR change detection method based on generalized Gamma distribution(GΓD)and Triplet Markov Fields(TMF).On the premise of obtaining the HLT difference map,this method uses TMF model to analyze the change information in the difference map.In the TMF model,GΓD is used as the likelihood probability distribution of HLT difference map to accurately describe the statistical characteristics of HLT difference map.TMF model is based on the neighborhood system and can consider the spatial correlation between neighborhood pixels.As a result,the TMF model can effectively suppress speckle noise,but it will lead to the lack of positioning details and make the boundary contour too smooth.Therefore,the auxiliary field is introduced into the TMF model to analyze the nonstationarity of the HLT difference map to describe the boundary information of the changing region,and combine it with the neighborhood information to improve the change positioning accuracy of the model.Finally,based on the Maximum Posterior Marginal(MPM)criterion,the TMF model iterates the marker field and auxiliary field for many times until the iterative accuracy is met,and the final marker field is the change detection result.Through the simulation experiment and analysis of four groups of measured data,the effectiveness of the algorithm in this chapter is proved.