SAR Image Denoising And Segmentation Based On Support Vector Machine

Synthetic aperture radar (SAR) is a kind of coherent microwave sensor. With its ability to image any targets on the earth, together with high resolution under nearly all weather conditions, it has been widely applied to national economic fields and military reconnaissance fields. In recent years, with the rapid development of SAR satellite astronomy technique, SAR image processing is becoming more and more important. Unfortunately, SAR images are complex in forming image, and always polluted by the speckle noise. For the above reason, SAR image processing is more difficult than common image processing.Support Vector Machine (SVM) is a new method of machine learning. It bases on the statistical learning theory, and can settle"small"example problem well. Because of its excellent learning ability, SVM has been applied to many fields. In this paper, based on studying basic theory and algorithm of SVM, we discuss support vector machine methods and their application in SAR image denoising and segmentation.In this paper, several typical approaches of image denoising are addressed. In view of the features of speckle noises in SAR image, we employ the regression capability offered by SVM network to construct a filter for image denoising, in which our feature selection and training data-set design enables the suppression of speckle noises. In the end, image denoising is performed by use of SVM filter. Experimental results demonstrate that the method works well for image denoising and it can also protect edge image information effectively.As regards image segmentation, we summarize many algorithms which have been presented in literature. Those algorithms are to some extent successful. In order to improve the segmentation performance of images corrupted by speckle noises, the fuzzy support vector machine method based on sigmoid kernel function is used in SAR image segmentation. The statistics features based on regional gray and the co-occurrence matrix of gray level are taken as training and testing data of FSVM. In the end, image segmentation is performed by using FSVM classifier. In order to demonstrate the effectiveness of the improved method, we use several division methods to process a SAR image and contrast the effect.The experimental results show that the method in this paper is feasible and accurate under noise distribution.