SAR Target Recognition Based On Dictionary Learning

Synthetic Aperture Radar(SAR) has been widely utilized in military and civilian fields owing to its capability of working under all weather and time conditions. With the development of SAR imaging technology, high-resolution SAR images and the large amounts of SAR data become available. SAR automatic target recognition has been a very hot research topic.This paper focuses on SAR target recognition study, including SAR image preprocessing, feature extraction and target recognition based on dictionary learning methods. The main contents of this paper are summarized as follows.1. The SAR image preprocessing and feature extraction are introduced. The relationship between the 3-D real scene and the corresponding 2-D SAR image is firstly discussed, and then the height feature extraction of vehicle target as well as the geometric feature extraction of ship target from the SAR image is studied. The ordinary order in which the preprocessing is followed by the feature extraction is not executed. Some prior knowledge easily obtained is used to judge the reliability of the results of the preprocessing and the feature extraction and to modify the results. Experiments verify the improvements of the accuracy of the extracted features.2. The SAR target recognition based on linear dictionary learning methods including K-SVD, D-KSVD and LC-KSVD is introduced. The comparison between K-SVD and K-means, the combination of unsupervised dictionary learning and supervised classification, and how to improve the discrimination of the dictionary are also discussed.3. Kernel method is introduced firstly, followed by the study of the SAR target recognition based on K-KSVD which is a non-linear dictionary learning method.