基于斯皮尔曼等级相关的单演谱成分选择及其在SAR目标识别中的应用

Selection of monogenic components based on Spearman rank correlation with application to SAR target recognition

特征提取是合成孔径雷达(synthetic aperture radar,SAR)目标识别中的关键因素之一.文中提出联合多层次单演谱特征的SAR目标识别方法,采用单演信号对原始SAR图像进行分解,获得不同层次的单演谱成分.基于斯皮尔曼等级相关分析分解的谱成分与原始SAR图像的相关性,设置相似度门限来选取若干具有较强鉴别力的谱成分.采用联合稀疏表示(joint sparse representation,JSR)对筛选得到的谱成分进行表征和分类,并基于MSTAR公开数据集在标准操作条件(standard operating conditions,SOC)和若干扩展操作条件下对多类地面车辆目标进行分类测试.实验结果表明:本文方法在SOC下对10类目标的平均识别率达到98.52%;对30°和45°俯仰角下的10类目标平均识别率分别为98.15%和72.06%;在噪声干扰条件下也可以保持良好的稳健性.综合对比,提出的方法相比现有几类SAR目标识别方法具有一定的性能优势.

Feature extraction is one of the key factors in synthetic aperture radar(SAR)target recognition.This paper proposes a SAR target recognition method by jointly using multi-level monogenic components.The monogenic signal is employed to decompose the original SAR images to obtain monogenic spectral components at different levels.Afterwards,the Spearman rank correlation is used to evaluate the similarities between different monogenic components and the original SAR image.A threshold is set to select those components with higher similarities with the original image.Then,the joint sparse representation is employed to classify the selected monogenic components for the classification.The proposed method is tested on the MSTAR public dataset under the standard operating condition(SOC)and several extended operating conditions(EOC)for the recognition task of a few ground vehicles.According to the experimental results,the proposed method could achieve an average recognition of 98.52%on ten classes of targets under SOC.The average recognition rates at 30°and 45°depression angles reach 98.15%and 72.06%,respectively.In addition,the proposed method could achieve good robustness under noise corruption.In comparison,the proposed method achieves superiority over several existing SAR target recognition methods.