相关性约束下SAR图像动态重构的目标识别方法  

Dynamic reconstruction of SAR images under correlation constraint with application to target recognition

作　　者：冯冬艳[1] 王海晖[2] Feng Dongyan;Wang Haihui(Department of Computer Engineering,ShanXi Polytechnic College,Taiyuan 030006,China;School of Computer Science & Technology,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]山西职业技术学院计算机工程系,太原030006 [2]武汉工程大学计算机科学与技术学院,武汉430205

出　　处：《电子测量与仪器学报》2019年第9期100-106,共7页Journal of Electronic Measurement and Instrumentation

摘　　要：研究合成孔径雷达(SAR)图像目标识别问题,提出在相关性约束下目标动态重构的方法。通过多层次的目标重构及决策融合提高整体识别性能。考虑到SAR图像的方位角敏感性,根据测试样本的估计方位角在测试样本中选取高相关性样本对其进行最优重构。通过调整相关性约束,获得测试样本在不同相关性下的重构误差,反映了训练样本与测试样本在不同层次的关联。采用线性加权融合的策略综合各个相关性下的重构误差,并根据最终融合的结果进行目标识别。基于MSTAR数据集设置了10类目标标准操作条件、俯仰角差异以及噪声干扰3种典型实验条件。结果表明,该方法在标准操作条件下对10类目标的平均识别率达到98.24%,在俯仰角差异及噪声干扰条件下的性能也优于对比方法,验证了方法的有效性。The synthetic aperture radar ( SAR) target recognition is researched and a method based on dynamic reconstruction of SAR image under correlation constraint is proposed. The target is reconstructed at multiple levels, whose results are fused to improve the overall recognition performance. Considering the azimuthal sensitivity of SAR images, the estimated azimuth of the test sample is used to select its highly correlated training samples, which are employed to best reconstruct the test sample. By modifying the correlation constraints, the reconstruction errors at different correlations can be obtained, which reflect the relation between the test sample and training samples at diflerent levels. Finally, the reconstruction errors from different correlations are combined by linear weighting fusion to determine the target label. Three typical experimental setups, i.e., 10-class recognition problem under the standard operating condition, depression angle variance, and noise corruption are designed based on the MSTAR dataset to test the performance of the proposed method. According to the experimental results, the proposed method achieves an average recognition rate of 98. 24% under the standard operating condition. In addition, its performance under depression angle variance and noise corruption is also superior over the compared methods. So, the effectiveness of the proposed method can be validated.

关 键 词：合成孔径雷达 目标识别 相关性约束 线性加权融合 

分 类 号：TN957[电子电信—信号与信息处理]