SAR Image Classification Based On Texture And Polarimetric Features

Synthetic Aperture Radar has been widely used in the fields of military and civil with its all-climate, day and night features and its images with meaningful polarimetric features and texture features. Interpretation results relating to more content are the basis for SAR application in many fields and SAR image classification is one of the key processes in the interpretation. Selecting appropriate features is very important to obtain high precision classification results.As one feature can only show targets’particular aspect information, the combination of various feature is beneficial to obtain accurate classification results just as it can better perform the subtle differences among targets.In view of texture features and polarimetric features have differences and complementarity in performing targets’feature, this paper combines texture features and polarization features to classify SAR image to make up for the deficiencies and limitations of image classification by using single feature.The main work and contributions are as follows:1)SAR image texture feature extraction is studied deeply. SAR image texture feature extraction methods are introduced based on gray level co-occurrence matrix and wavelet transform. The determination of parameters is analyzed in feature extraction process. Target distinguishing ability in texture feature is analyzed by using mean value, histogram, correlation coefficient, and class separation methods.2) The scattering mechanism of typical targets is analyzed deeply based on polsar decomposition, polarization synthesis and the covariance matrix elements. Different manifestations of typical targets in polarization parameters are studied.3) Target recognition based on support vector machine principle is summarized. Method of SAR image multi-class classification based on support vector machine is given.4) PolSAR image classification based on support vector machine is researched, and SAR image classification experiments in GenHe are conducted based on suitable texture and polarimetric features. Then accuracy assessment and comparative analysis of classification results are given, especially GLCM, wavelet texture impaction on classification accuracy is studied.