Ship Target Recognition Based On ISAR Images

Since World War II, the radar system has become more and more important in the modern war. Because of its ability of acquiring considerable information, Inverse Syn-thetic Aperture Radar(ISAR) is a good application environment for Automatic Target Recognition(ATR) techniques. As coast-based ISAR has the ability of long range imaging to ship targets and of obtaining the shape and structure features, the ship ATR technique based on ISAR images has a wide development prospect and huge application value. This paper carries research about the ship ATR technique based on the experiments of real ISAR data. The main contents of the paper are detailed as follows:First of all, in order to suppress the influence of the typical noise in ship ISAR images and to solve the problem of imperfect structures, this paper proposes a whole image pre-processing scheme including some image segmentation and morphological methods. The taget images will be clear and perfect without any noise after being pre-processed.Second, this paper investigates the extraction methods of some shape features. The robustness of these features under the circumstances of fluctuating cross-range scaling results and of different distributions of samples are tested on the real ISAR data. The results show that these features can be applied in the real application.Besides, this paper proposes several novel automatic segmentation algorithms to the superstructure curves of ship targets. The Mean-Gradient algorithm segments the curve at the visually most reasonable positions. The Mean-Comparation algorithm and the Clus-ter algorithm achieve the optimal segmentation by making the heigt distributions in each segment closer. The Turning-Point-based-Cluster algorithm combines the two ideas. The results of real ISAR data experiments testify that our automatic sementation alogrithms improve the differential ability of the superstructure code feature.Finally, this paper combines the fuzzy set theory with the superstructure code and propose a novel superstructure-based fuzzy recognition technique. By using a fuzzy set on the universal set of all codes to represent the target, this technique enhances the de-scriptive ability of the old superstructure code feature and utilizes the maximum clossness principle to classify the targets. Based on the segment comparative mean feature, the tech-nique uses a two-step method to compute the memberships to the target of all codes. The target is then represented as a fuzzy set and classified as the class which has the largest closeness. The recognition results of three closeness definitions-Hamming closeness, Euclidean closeness and lattice closeness-are compared in the real ISAR data experi-ments, while the results also prove that the fuzzy recognition technique proposed in this paper improves the classification rates of other published superstructure-based techniques.