带鉴别分析的多视角SAR图像联合决策及目标识别

Joint Decision of Multi-view SAR Images with Discrimination Analysis with Application to SAR ATR

提出多视角SAR图像联合决策的目标识别方法。多视角SAR图像对于描述目标具有互补性,因此可以为目标识别提供更丰富的鉴别信息。考虑到参与融合的部分视角可能对最终决策融合的贡献较弱。本文在多视角SAR图像的决策融合中引入鉴别分析。对于单个视角的SAR图像,首先基于支持向量机分类器进行分类,根据输出的决策值定义当前视角的可靠度。然后,采用门限判决的方式选取可靠性较强的视角,并基于贝叶斯决策融合对选取视角的分类结果进行联合决策从而判定目标类别。基于MSTAR数据集进行目标识别实验验证提出方法的有效性。

This paper proposes a synthetic aperture radar( SAR) target recognition method based on joint decision of multi-view SAR images. The multi-view SAR images complement each other as for describing the target thus providing more discriminability for target recognition. Considering that some views may not be discriminative enough for decision fusion,the discrimination analysis is proposed in this paper. For each view,it is first classified by Support Vector Machines( SVM). Afterwards,a reliability value is defined based on the output values from SVM. A thresholding method is used to select those views with higher reliabilities,which will be fused based on the Bayesian decision fusion for robust target recognition. Experiments are conducted on the Moving and Stationary Target Acquisition and Recognition( MSTAR) dataset to validate the effectiveness of the proposed method.