SAR Image Classification Algorithm Based on Improved SIFT Features

With the advancement and development of Synthetic Aperture Radar (SAR) technology, target classification and interpretation based on SAR image data has become one of a hot research directions for domestic and foreign scholars. Most early SAR image interpretation work relies on overall features or original intensity values, such as template-based classification methods, but such methods are very sensitive to local changes. So, it is necessary to find a method that can effectively extract the local features of the image. In order to improve the recognition performance of vehicle targets in SAR images, this paper summarizes a method suitable for vehicle target recognition in SAR images by testing key points and local descriptors. First, In the key-point detection module, Difference of Gaussian, Harris corner detection, and Features From Accelerated Segment Test (FAST) algorithms are used to generate key-point. In the descriptor extraction module, different modes are used for descriptor gradient statistics to extract Scale Invariant Feature Transform (SIFT) descriptors; secondly, the SIFT feature points are clustered using K-Nearest Neighbor (KNN) to generate a dictionary (Bag of words), then use the Spatial Pyramid matching (SPM) algorithm to generate the feature vector of the picture and use the Support Vector Machines (SVM) classifier for classification; finally, the detection performance of these algorithms is quantitatively analyzed and the optimal performance module is selected to combine modules as the final classification algorithm. It was verified on the Moving and Stationary Target Acquisition and Recognition (MSTAR) data set, and the results showed that the performance of the algorithm has been greatly improved than before.