SAR Image Classification Based On Multi-Feature And Composite Kernels

Synthetic Aperture Radar(SAR)can acquire high quality images of different land coverage at any time in any weather conditions.Therefore,SAR has been successfully applied in many fields,such as military target detection and identification,environmental monitoring and land topographic mapping.However,due to the particularity of the imaging principle of the SAR system itself,the SAR image has inherent multiplicative speckle noise,which brings great difficulties to the interpretation of SAR images.In recent years,SAR image classification has received extensive attention as a key part of the image processing field,and has been an important part of SAR image analysis.In this paper,the multi-feature and composite kernels are organically combined to carry out in-depth analysis and research on SAR image classification.The specific research work includes:1)SAR image classification algorithm based on multi-feature and support vector machine(SVM).Aiming at the problem that the single feature of SAR image cannot fully describe the target and extract rich information,a multi-feature extraction method combining gray level co-occurrence matrix(GLCM)and multi-level local pattern histogram(MLPH)is proposed.This method uses gray level co-occurrence matrix to describe the spatial feature of SAR image,uses multi-level local pattern histogram to describe the structure feature of SAR image,and then combines the two kinds of feature information,on this basis,the support vector machine is used to classify the SAR image,and the effectiveness of the proposed algorithm is proved by the experiments of synthetic and real SAR images.2)SAR image classification algorithm based on superpixel and composite kernels(CKs).Aiming at the problem that the classification of SAR images by a single kernel is not accurate enough,a method of applying composite kernels(CKs)in support vector machine is proposed.The method first generates context information from superpixel,then defines the context kernel through the radial basis kernel function(RBF),and then fuses the context kernel with the radial basis kernel function to form a composite kernels(CKs).Finally,the classification of pixels is realized by CKs-based support vector machine(SVM)combined with multi-feature extraction.The outcome display that the method is not only highly accurate,but also can deal with non-homogeneous regions and weak edge classification problems.