Sar Image Surface Features Based On Svm Classification

Synthetic Aperture Radar (SAR) owns high resolution which relies on the modeling large-aperture antenna by moving carrier to obtain the radar imaging data of high orientation resolution. Compared with optical sensor, SAR can work all-weather and all-time because of using microwave remote sensing. Under this condition SAR owns some priorities such as multi-polarization, changeable angle of view and penetrability. Resolution of SAR is improving as its technique develops. Recently its resolution has approached or already exceeded the resolution of optical imaging. And the usefulness of SAR is a hot point. Although advantages SAR owns its information processing technique especially the technique of massive data real-time interpretation is still on the initial Stage. So how to interpret the data of SAR quickly and precisely has become a universal problem.The theme of this paper is the extension of the traditional ATC (Automate Terrain Classification) technology. On the one hand, it can act as the front part of the SAR interpretation system, capture the ROI (Region of interest) with level information, instead of the target detection and discrimination, provide potential target chips for the target recognition, on the other hand, it can directly provide important parameters for manual interpretation, and establish independent assistant interpretation system. So, the research of this paper is a very practical and pivotal part according to the present development level of the SAR image interpretation technology.The study of this paper is the application of Support Vector Machine (SVM) on terrain classification of SAR image. Around this, firstly, we discuss the characters and produced mechanism of speckle noise, pose a new de-noise processing measure; Secondly, we pull-in a way of feature extraction, depend on not only gray feature after de-noising, but also texture feature based on Gray-level Co-occurrence Matrix in the space domain and Wavelet packet Transform in frequency domain. Thirdly, we systemically introduce a powerful data analysis tool—SVM. By the study of the theory of SVM and the analysis the research of SVM classified arithmetic, we design a classifier of Least Square Support Vector Machine (LSSVM) to solve the parameter select and improve the speed; In the end, we test the advanced algorithms by the real MSTAR SAR data. Applying the designed classifier to SAR images, the results confirm the feasibility and accuracy of the algorithm.