基于BVMD特征决策融合的SAR目标识别方法  

SAR Target Recognition Based on Decision Fusion of BVMD Features

作　　者：莫海宁[1] 钟友坤 MO Haining;ZHONG Youkun(HTC VIVEDU School of Technology,Guangxi University of Science and Technology,Liuzhou 545006,Guangxi,China;Physics and Mechanical&Electrical Engineering School,Hechi University,Yizhou 546300,Guangxi,China)

机构地区：[1]广西科技大学宏达威爱科技学院,广西柳州545006 [2]河池学院物理与机电工程学院,广西宜州546300

出　　处：《电子信息对抗技术》2022年第5期40-44,50,共6页Electronic Information Warfare Technology

摘　　要：针对合成孔径雷达(Synthetic Aperture Radar,SAR)目标识别问题,将二维变分模态分解(Bidimensional Variational Mode Decomposition,BVMD)用于特征提取进而进行分类决策。BVMD将原始SAR图像分解为若干模态,实现对目标特性的层次化细致描述。采用稀疏表示分类(Sparse Representation-based Classification,SRC)分别对各个模态进行分类进而根据输出的重构误差定义各个模态的分类置信度。通过门限法选取若干具有高置信度的模态进行决策融合从而判定测试样本的类别。实验中,在MSTAR数据集上对提出方法进行验证。通过在标准操作条件和扩展操作条件下与几类现有方法进行对比,结果证明该方法是有效和稳健的。For the problem of Synthetic Aperture Radar(SAR)target recognition,the Bidimensional Variational Mode Decomposition(BVMD)is applied to feature extraction and classification.BVMD is employed to decompose original SAR images into several modes,which provide multi-level and detailed descriptions of the targets.Each mode is classified by the Sparse Representation-based Classification(SRC)and a reliability level is defined based on the reconstruction errors.A thresholding algorithm is further employed to select several modes with high reliabilities.A decision fusion is performed afterwards to combine the results from the selected modes to determine the target label.In the experiments,the proposed method is validated on the MSTAR dataset.By comparison with some current SAR target recognition methods under the standard operating condition and extended operating conditions,the effectiveness and robustness of the proposed method are demonstrated with the results.

关 键 词：合成孔径雷达 目标识别 二维变分模态分解 稀疏表示分类 决策融合 

分 类 号：TN957.52[电子电信—信号与信息处理]