SAR Target Discrimination In The Complex Scene

Using image processing and pattern recognition techniques, Synthetic aperture radar(SAR) Automatic Target Recognition(ATR) system can realize automatic target recognition, which is widely studied all over the world. SAR ATR mainly includes three stages which are SAR target detection, SAR target discrimination and SAR target recognition. The second part of this system is used to reject the clutters in the candidate target chips. The conventional SAR target discrimination methods can obtain good results in simple backgrounds but do not perform good enough in complicated backgrounds. Therefore this paper focuses on the study of target discrimination to solve the problems above. The main contents of this paper are described as follows:1. The first part in this paper mainly describes the research background and the significance of the SAR ATR and introduces the development of target discrimination.2. The second part in this paper introduces the two-stage target discrimination algorithm based on the classification of clutters. Clutters are classified into two parts, the natural ones and the man-made ones, to reject clutters viatwo steps. The first step is conducted to reject the natural clutters, and the second step the man-made ones. According to the experimental results, this method shows better performance than the conventional algorithm.3. The target discrimination based on histogram of oriented gradients(HOG) featureis described in the third part. Firstly, gradient amplitude images and gradient orientation images should be extracted using edge detection to draw a histogram, which determines the main orientation in the image; Secondly, the gradient amplitude images and gradient orientation images are rotated according to the main orientation above.Then the rectangle with optimum matching is obtained by conducting template matching operation on the rotated gradient amplitude images; Thirdly, HOG features are extracted; Finally, support vector machine(SVM) is used to accomplish target discrimination in complicated backgrounds.4. A SAR target discrimination method based on the bag of words model is introduced. Firstly, local areas in the image are extracted and the extracted aligned HOG features are used as local descriptors. Secondly, the visual codebook is constructed via local descriptors of training samples; Thirdly, the frequency of every word both in the training samples and the testing samples is calculated, which forms the histogram for each image, and this histogram is considered as the new features for the target discrimination; Finally, support vector machine(SVM) is used to discriminate the SAR target from man-made clutters in complicated backgrounds.