Target Recognition in SAR Images Based on Information-Decoupled Representation

摘要: This paper proposes an automatic target recognition (ATR) method for synthetic aperture radar (SAR) images based on information-decoupled representation. A typical SAR image of a ground can be divided into three parts: region, shadow and background. From the aspect recognition, region contain discriminative information. However, they also include some confusing information because similarities different targets. The background mainly contains redundant information, which has little contribution to recognition. Because segmentation may impair in relatively simpler is performed separate decoupling. Then, representations are generated, i.e., image, original image. retained represents coupling backscattering generated classified using sparse representation-based classification (SRC). their results combined by score-level fusion not used its lower discriminability possible errors. To evaluate performance proposed method, extensive experiments conducted Moving Stationary Target Acquisition Recognition (MSTAR) dataset under both standard operating condition (SOC) various extended conditions (EOCs). correctly classify 10 classes targets with percentage correct (PCC) 94.88% SOC. With PCCs 93.15% 75.03% configuration variance 45° depression angle, respectively, superiority demonstrated comparison other methods. robustness uniform nonuniform errors validated over 93%. Moreover, maximum average precision 0.9580, more effective than reference methods outlier rejection.