A Novel Approach For Target Recognition Using Feature Extraction Over RCS Data

The rises of modern high resolution radar makes the automatic target recognition (ATR) available, by analysis of the backscattering of target, the position, the velocity, the elevation angle, and more detail information, such as the geometry feature of the target can be acquired, but only by further data processing can make the ATR be effective.In this paper, a novel approach for target recognition and size determination by using texture feature extraction over RCS (Radar Cross Section) is proposed firstly. Here, the frequency-angle RCS data sets are normalized for every observation angle as intensity images with 256 gray levels, and different target's RCS data sets can be transformed to gray-level images with different textures. This phenomenon implicates that the identity textures of each image can be used to recognize the corresponding target and estimate the size of the unknown target. Here, we chose the gray level co-occurrence matrix for texture feature extraction. The observed results of recognition show that the proposed approach is of greater perspective for target discrimination than many other available techniques. By applying advanced classifier and texture feature analysis, the proposed method can be used in ATR.Here, we first discuss the importance and background of radar target recognition, the main techniques and achievements are also listed. Then,brief introductions of the essence concepts in radar research, which are closely related this study, are given one by one. As these points are the basic theories for target recognition, and mentioned while introduce the novel approach. Among the various approaches for radar target recognition, only 4 ones are presented in some detail. As the analysis of full-aspect & monofrequency RCS data for size estimation, 1-D Range Profile,2-D SAR/ISAR Radar Imaging and the Scatter Center Reconstruction Based on ICA are the enlightenments of this study, and the new approach introduced can partly overcome the difficulties in the traditional methods, recognition results of simulation data are also presented. After that, the applicability of the new method, comparison with traditional methods and the outlook are presented.