基于多尺度特征的高分辨率SAR图像机动目标识别

针对客观世界本身的多尺度特性及多尺度分析方法的有效性，提出了能够体现目标宏观和微观固有多尺度信息的SAR图像目标识别方法。该方法通过引入目标先验信息分割出反映目标形状的区域；并根据SAR图像目标对Gabor滤波器组的不同响应，利用多尺度信息及尺度间的相依性提取新的多尺度特征；最后按照方位信息建立多尺度特征库用以指导目标识别。通过MSTAR的实测SAR目标数据集进行了多尺度特征建立和目标识别测试，试验结果表明了该方法的有效性。

Maneuvering target recognition of high resolution SAR images based on multi-scale feature

The real-world itself has multi-scales characters, and objects only make sense in special scales. A method of targets recognition in SAR images based on multi-scale information is proposed, in which objects’ macroscopic and microcosmic characters are well presented by multi-scale analysis. The SAR images are segmented by introducing a priori knowledge in order to reflect the targets’ shape information. According to the different response characters of targets corresponding multi-scales Gabor filters, multi-scale information and the pertinence between multi-scales are adopted to extract new multi-scale features. Finally, the new multi-scale features database is established according to the azimuth angles of targets to guide targets recognition. Multi-scale features extraction and targets recognition are tested on MSTAR SAR data. The experimental results show the validity of this method.