Feature Extraction Based On Affine Projection Matrix And Aerial Targets ISAR Image

Since the beginning of the50’s in last century, the capability of obtaininginformation is more and more important to the war. In this context, RadarAutomatic Target Recognition (ATR) as a new field appeared. Synthetic ApertureRadar (SAR) and Inverse Synthetic Aperture Radar (ISAR) development providesvery strong support for the development of ATR. They can achieve remote imagesand different angles’ images.They also can provide structural features of the object.So it is conducive to the realization of the target feature extraction and recognition.This paper based on Aircraft target’s ISAR simulation, focuses on the research ofthe ISAR image feature extraction and target recognition. The main contents areorganized as follows:Firstly, We can establish three kinds of aircraft target imaging model based onISAR imaging principle. Then we can use Range-Doppler algorithm to obtain therequired database object recognition. Which based on Three-Dimensional (3D)scattering point model of aircraft target.Secondly, feature extraction is based on ISAR images. To eliminate the targetand the imaging plane affine transformation as a theoretical basis, We propose anew approach: Affine projection matrix In order to compare the characteristics ofstability and efficiencyˋThe paper also uses other two feature extraction methods The first one is Hu invariant momentsˋHu invariant moments is a very classicimage feature extraction method, which proposed seven kinds of characteristicstability is very good; the second one is the wavelet singular value, WaveletSingular value is nearly two years the proposed new method, which uses theprinciple of wavelet analysis to extract two-dimensional wavelet transform of thefour sub-graphs singular value feature, it is a new wavelet coefficients fusionfeature extraction method.Finally, to target identification phase, the paper selected two classicclassification algorithms: Nearest neighbor method and neural networks. First, wedescribes the two basic principles of classification algorithms, and analyzes theerror rate of the nearest neighbor method of upper and lower bounds. Then usingthe above three features to see different situations target recognition rate. We cananalyze target recognition rate to compare the performance indicators betweenaffine projection matrix and the other two feature.