Research On Change Detection Method In Multi-Temporal Polarimetric SAR Imagery

As a technology of identifying the changes that have occurred on the earth's surface by multi-temporal images acquired in the same geographical area at different times,change detection play an important role in study on land-use dynamics,environmental monitoring,disaster warning and et al.Owing to the ability of obtaining land cover information in short time,remote sensing techniques,which can save manual work and material resources,have been successfully used in change detection.Polarimetric Synthetic Aperture Radar(PolSAR)images contain both phase and amplitude information from the radar returns transmitted in two different polarizations,and use more scattering information.Moreover,compared with single-channel SAR images,it has a great advantage in the analysis and extraction of target.With the rapid development of airborne and spaceborne SAR systems,more and more PolSAR data are available.How to use of polarization information in change detection is the frontier problem of PolSAR application,and has important research signification and great application value.In this thesis,the multi-temporal PolSAR data of spaceborne RADARSAT-2 and airborne UAVSAR are taken as the research object.Based on the full use of polarization information,the unsupervised and supervised change detection methods are studied in-depth.The main contributions of this thesis are presented as follows:(1)We introduced and summarized the existing multi-temporal SAR images change detection algorithms,and divided current methods into unsupervised and supervised approaches.On the one hand,we introduced the unsupervised methods based on the procedure of preprocessing,generating the comparison image,choosing the threshold and accuracy assessment.On the other hand,we introduced the supervised methods based on the flow of preprocessing,classification,comparing the classification results and accuracy assessment.(2)In the aspect of comparison image extraction,in order to solve the problems that generating the comparison image mainly focus on single-polarized and multi-polarized SAR and used few polarimetric information,we concluded the advantage of PolSAR in change detection.Compared with single-channel SAR images,PolSAR sensors acquire phase and amplitude information in different polarizations,and thus offer more scattering information.Moreover,using all elements of the Coherence/Covariance matrix,we developed the test statistics to generate the difference image and improved the discrimination of unchanged and changed regions.The experimental results show that the proposed method is effective and highly accurate for the comparison image,and less affect for the disturbing factors.(3)In the aspect of choosing the threshold in unsupervised change detection,we concluded the some traditional methods.Based on the statistical distribution characteristics,we used the generalized Gaussian model to simulate the comparison image,and proposed a method based on Improved Kittler and Illingworth(IK&I)to automatically determine the threshold.The experimental results show that this method can effectively improve the accuracy of change detection.(4)In the aspect of supervised change detection based on multi-temporal PolSAR images,supervised methods are easily affected the classification cumulative error,and unsupervised methods cannot detect the type of land cover change.In order to solve these problems above,a joint classification comparsion(JCC)strategy was proposed.Using the least variance of intensity,we decided the sequence of classification.Moreover,the similarity measure based on the full polarimetric information was proposed to control the joint classification.In this thesis,the method of JCC based on similarity measure is proposed and the framework is constructed.The experimental results show that the proposed method obtain the high accuracy region of change detection with less training samples.(5)In the monitoring of wetland water level changes,the current time-series SAR image detection methods for wetland water level changes are summarized.In the change detection applications,interferometric SAR(InSAR)relies on the coherence of time-series SAR images and PolSAR cannot detect the change degree.In order to solve these problems,we combined PolSAR and InSAR technologies to monitor the wetland water level over a large area.We used the unsupervised method of change detection to search the unchanged region.Using polarimetric decomposition,copolarized phase difference and Coherence map,the scattering characteristics of unchanged region were identified to choose the training samples.Moreover,these training samples were used in JCC to obtain the land cover change.The change of land cover corresponded with the water change.Combined with the technology of PolSAR and InSAR,the water-level change in wetland can be estimated.