Change Detection In PolSAR Images Based On Difference Analysis And Level-set Conditional Triplet Markov Field

Synthetic Aperture Radar(SAR)has been widely used in the field of remote sensing due to its excellent performance in earth observation.Polarimetric Synthetic Aperture Radar(Pol SAR),as an extension of SAR,is capable of alternately transmitting and receiving signals in a multi-polarization manner to further acquire scattering information of feature targets,providing a solid data basis for remote sensing tasks such as segmentation,classification,water body extraction,target detection and change detection.As the research priority of remote sensing image analysis and interpretation,change detection plays an important role in land cover monitoring,offshore oil spill detection,disaster situation analysis and so on.The imaging mechanism of SAR leads to the suppression of speckle noise as a common problem in many SAR image processing techniques.In change detection,the speckle noise can degrade the image quality of the difference map and ruin the texture information,thus affecting the effect and precision of change detection.Therefore,how to maintain the texture information and accomplish precise localization while suppressing the noise becomes a research hotspot also a pinch point.In this thesis,we will improve the traditional polarimetric SAR change detection method in two perspectives,that is difference information extraction and change information analysis.Aiming at the edge fracture and aliasing problems of the change map generated by above method,Chapter 4 takes the difference information extracted by the method in Chapter 3 as a priori,and proposes the polarization SAR image change based on the level set condition three fields(LS-CTMF)Detection algorithm.The method first uses the Equivalent Number of Looks(ENL)estimation method based on Trace Moment-based Estimator(TME)to calculate the ENL map,and selects a threshold to divide it into heterogeneous and homogeneous regions.After that,the Maximum Hotelling-Lawley Trace(MHLT)is introduced as the basic difference calculation operator.In heterogeneous regions,the HLT operator is combined with the neighborhood hypergraph model to form the hypergraph model-based HLT operator(HG-HLT),using to analysis and extract the difference information at texture detail region,which is beneficial to maintain texture change information.In homogeneous regions,the HLT operator is combined with the improved neighborhood ratio model to form improved neighborhood ratio model-based HLT operator(IN-HLT),using to extract the difference information of smooth blocks,where effectively suppress the speckle noise.Finally,the two parts of difference information are fused and the Expectation Maximization(EM)algorithm is used to generate the change map.The feasibility and advantages of the algorithm in this chapter are verified by four sample sets of Gaofen-3 polarimetric SAR data.Aiming at the problem of edge fracture and jaggedness in the change map generated by the above method,Chapter 4 proposes a polarimetric SAR image change detection algorithm based on Level-Set Conditional Triplet Markov Field(LS-CTMF)with the change detection results of the Chapter 3 method as a priori information.The algorithm first constructs the difference map after noise suppression using the algorithm in Chapter 3,and then analyzes the change information on the difference map using Conditional Triple Field(CTMF)to optimize the details of the change region.The modeling process introduces an auxiliary field initialized with zero level set to provide a priori information of the edges,while a level-set constraint is imposed to limit the update range of the auxiliary field during the update process,which ensures the continuity of the auxiliary field at the edges and better locates the edges of the change region.The feasibility and advantages of the algorithm in this chapter are verified by four sample sets of Gaofen-3 polarimetric SAR data.