The Research For Sar Image Classification & Small Target Detection

Over the past few years, the development of spaceborne and airborne Synthetic Aperture Radar (SAR) systems have experienced a rapid growth, for its all-weather and high resolution capabilities . Computer science and modern signal processing methods make it possible to apply technology of SAR in many areas, such as remote sensing and mapping, exploration of terrestrial resources, catastrophe forecast, military maneuvering and other civilian applications.Image classification is intended to determine the closest match between a data sample and a set of samples representing the difference classes in the image. In the interpretation of SAR image, texture provides important information, in addition to image gray level or backscatter coefficient alone. Studies have shown that classification based one texture features can improve the accuracy of the interpretation.In this thesis, first, author investigates the performance of texture features derived from the gray-level co-occurrence matrix, then uses K-means clustering to achieve the pre-classification of SAR image, at the same time, discusses the difference between variant texture features in SAR image classification. Secondly author weights texture features by reciprocal of every class' variance. In the end author finishes post-classification of the weighted image by K-means clustering again and the classification accuracy is increased significantly.In the latter part of this thesis, in order to achieve small target detection of SAR image, author uses Order Morphology and High Pass Filter methods to pre-process image, meanwhile discusses and compares contributions of these two methods to such aim. Then, depending on its geometric characters, linear point array targets are detected from the image processed by High Pass Filter method.