针对合成孔径雷达(SAR)图像目标识别中的特征提取和分类决策问题,提出了基于多模态聚类和和决策融合的方法。采用二维经验模态分解对SAR 图像进行分解,获得多个模态表征结果,实现目标特性的多层次描述。基于多个模态的内在相关性进行聚类,获得若干模态子集。基于联合稀疏表示分别对各个模态子集进行分类,获得相应的重构误差矢量。利用线性加权融合对各个模态子集的结果进行融合,最终对测试样本的类别进行决策。基于MSTAR 数据集开展实验,结果表明,该方法在标准操作条件下的有效性和扩展操作条件下的稳健性。

For the feature extraction and classification problems in synthetic aperture radar (SAR) target recognition, a method based on clustering of multiple modes and decision fusion is proposed. The bidimensional empirical mode decomposition is employed to decompose SAR images to obtain multiple modes, which describe the target characteristics from different aspects. The inner correlations among all the modes are analyzed to cluster them into several subsets. The joint sparse representation issued to classify each subset to get the corresponding reconstruction error vector. The linear weighing fusion is used to fuse the results from different subsets and finally decide the target label. Experiments are conducted on the MSTAR dataset and the results show the effectiveness under SOC and robustness under EOCs.

TN753

江苏省高等学校自然科学研究面上项目(18KJD520001)