Research On Classification And Change Detection In Polarimetric SAR Imagery

Polarimetric Synthetic Aperture Radar(PolSAR),by receiving electromagnetic waves with different polarization combinations of ground objects,acquires more abundant scattering information than Synthetic Aperture Radar(SAR),which brings new opportunities and challenges for the development of PolSAR classification and PolSAR change detection.It is widely used in many fields,such as urban planning,geological disaster detection and military target strike.This thesis summarizes the methods of PolSAR classification and PolSAR change detection at present.This thesis studies based on Gaofen-3 data,combined image statistical characteristics and physical scattering mechanism for PolSAR classification.Two or more different temporal polarimetric images of the same scene are detected and analyzed based on Sentinel-1 data.The main research contents are as follows:(1)In order to solve the problem of the unreasonable classification of ground objects by traditional unsupervised classification methods,this paper base on a multiple component scattering model(MCSM)proposes a MCSM-K-Wishart classification method.Firstly,deorientation compensation is used to improve the problem that the volume scattering contribution is over-estimation in MCSM decomposition.Then,Isodata clustering method and K-Wishart method were introduced to overcome the disadvantages of arbitrary classification and unsuitable for Gaussian distribution scenes.Finally,gets the initial PolSAR classification results.(2)The scattering characteristics of some ground objects are similar,which leads to indistinguishable and missed detection.To solve these problems,a method to identify some ground objects based on the inherent physical scattering was proposed.In this method,the coefficient of variation was used to distinguish water and bare soil,and the texture features were used to improve the unclear contour of some buildings.Finally,the final PolSAR classification results are optimized by the method of mathematical morphology.(3)Regarding that the complexity of the actual scene and serious speckle noise,this research introduced a dual-temporal PolSAR change detection method based on BM3 D joint weighting.This method focuses on selecting the difference degree model that meets the actual scene by adjusting the weight coefficients,but the inherent speckle noise of the PolSAR image and the difference degree image will affect the accuracy of the change detection.This article uses the BM3 D algorithm to reduce noise,and uses the generalized Gaussian KI to automatically determine the threshold.The effectiveness of this method is verified by the measured data processing.(4)Considering that the existing bi-temporal PolSAR change detection cannot reflect the dynamic change trend of ground objects from the time dimension,this paper proposes a new method of time-series PolSAR change detection.By defining the similarity change detection matrix and statistical hypothesis testing,the change detection analysis is carried out on the long-time PolSAR image sequence pixel by pixel.Obtain change time nodes in the time domain and extract changing pixels in the space domain.The experimental results show that the method can effectively improve the accuracy of change detection in multi-temporal PolSAR images.