A Region Matching Approach Based on 3-D Scattering Center Model With Application to SAR Target Recognition

This paper proposes a region matching approach based on 3-D scattering center model with application to synthetic aperture radar (SAR) automatic target recognition (ATR). Each of the model scattering centers is represented by a binary region obtained by segmenting its predicted image using a global threshold. Afterwards, a region-to-region matching algorithm is performed to match the individual predicted regions with the extracted target region from the test image. Three scores are designed to comprehensively evaluate the matching result, which are linearly fused as an overall similarity based on their sparabilities as for discriminating different targets. In the classification stage, a coarse-to-fine region matching is performed and the test image is decided to be the class with the maximum similarity. Experiments are conducted on the simulated data by the electromagnetic code. The results demonstrate that the proposed method outperforms several template-based and model-based SAR ATR methods under the standard operating condition. The superiority of the proposed method becomes more remarkable under various extended operating conditions, including noise corruption, resolution variance, and partial occlusion.