Research On Automatic Recognition In Unmanned Aerial Vehicle SAR Images

With technology development of the Unmanned Aerial Vehicle (UAV), Synthetic Aperture Radar (SAR), as UAV’s mainly radar, is the main object of study at home and abroad, especially the technology research of SAR images Automatic Target Recognition (ATR) has been widespread concerned. In this paper we present SAR technology with the database provided by United States AFRL /DARPA MSTAR, aimed at the key technologies in target of SAR ATR feature extraction and target classification involved launched research at home and abroad,according to SAR ATR of three steps,including the target detection, identification and classification.We research the image enhancement, target method of feature extraction and target classification method and so on ,combined with the development trend of the SAR ATR. We did the simulation experiment in matlab7.0 platform.Firstly, according to the general steps of SAR ATR, we study the SAR data of MSTAR and the SAR image noise situation, introduce the K-distribution, Gamma distribution and distribution of Weibull distribution model noise. Through the research of noise, we prepare for the subsequent work of classification.Then, through the study of enhancement on SAR image, we propose an improved amplitude domain regularization method according to amplitude image enhancement of regional characteristics requirement. The experimental results can be apparent that the improved regularization method can distinguish obvious regions into target, shadow and background. Then CFAR detection algorithm is studied, the main research of double parameters CFAR detection algorithm, and through the experimental effect validation.Next, we study the target feature extraction in discrimination of SAR images. We propose a feature extraction method according to target type classification method of target features, with the view of the present few articles on feature extraction method according to target type classification problem, they are target object feature of vehicles, buildings, roads & Bridges and ships ; we also establish the SAR image targets of 2-d classification framework. Then we emphasis on the research of Hu seven invariant moments on target and shadow, proposed by fourteen Hu moment invariants are composed of characteristic vector and characteristics for the following classification process.Finally, we study the commonly used method in SAR image classification and classification model, focused on the FMM classification method and LVQ classification method, we propose a kind of method combined three kinds of classification methods, it composites the results of FMM classification , LVQ1 classification and LVQ2 classification with voting, and achieved good classification effect.