考察独立性和相关性的多视角SAR图像目标识别方法  

A Target Recognition Method for Multi-view SAR Images in Consideration of Independency and Correlation

作　　者：陈婕[1] CHEN Jie(Institute of Information Technology Guilin University of Electronic Technology,Guilin 541004 China)

机构地区：[1]桂林电子科技大学信息科技学院,广西桂林541004

出　　处：《电光与控制》2020年第3期89-93,114,共6页Electronics Optics & Control

摘　　要：提出了一种考察独立性和相关性的合成孔径雷达(SAR)图像目标识别方法。由于SAR图像的方位角敏感性,参与识别的多视角SAR图像之间的关联性不够稳定。首先,基于图像相关对多视角SAR图像进行聚类,获得具有较强内在关联的多个视角集。然后,对于每一个视角集,分别采用联合稀疏表示对其进行联合重构,获得高精度的重构误差。最终,采用线性加权的方法融合各个视角集合的重构误差并根据融合误差判定目标类别。基于MSTAR数据集进行了实验测试,结果表明了提出方法的有效性。A Synthetic Aperture Radar(SAR)target recognition method in multi-view SAR images is proposed in consideration of both independency and correlation.Due to the azimuth sensitivity of SAR images the correlations among the multi-view SAR images for recognition are not stable enough.Therefore the multi-view SAR images are first clustered into several sets based on image correlation.Images in each set share relatively high correlations.Afterwards Joint Sparse Representation(JSR)is employed to jointly reconstruct the multi-view SAR images in each view set and produces the reconstruction errors with high precision.Finally the linear weighting strategy is used to fuse the reconstruction errors from different view sets to determine the target label.Experiment tests are carried out on the MSTAR dataset.The results prove the effectiveness of the proposed method.

关 键 词：合成孔径雷达 目标识别 多视角 独立性 相关性 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]