基于峰值点的目标对准在MSTAR SAR图像分类中的应用

模板匹配的分类方法是合成孔径雷达目标识别的重要方法。研究了MSTAR SAR图像,指出图像中的目标峰值点本质上对应于实际物体的散射中心,并且在一定方位区间中具有稳定性。假设目标存在位置偏移的基础上,提出先利用目标峰值点对准目标,再生成模板和分类的思想。分析了方位区间中基准图像的选取及对准过程对模板生成和目标分类的影响程度。最后研究了同类变形目标间的分类,提出了峰值点特征增强成像、预分类、目标对准和正式分类的四步算法。结果证明,所提算法具有较高的分类精度和效率。

Peak Based Target Alignment and Its Application to Classification of MSTAR SAR Images

The template-based classification is popular with synthetic aperture radar automatic target recognition.It is pointed out that the SAR trget peak is Corresponding to the Scattering Center of a target,and is invarant in an azimuth interval in nature.Based on the assumption that the target position shift existed in SAR images obtained in real battlefield,a scheme of target alignment based on peak is proposed.This scheme is to use the target peak of the image to align the target firstly,then to do template training and target classification.The selection of the reference image in azimuth interval is discussed,and the influence of the alignment on template training and target classification is discussed too.At last,the classification for a target with variant configurations is described which is consisted of the target extraction based on the feature enhanced SAR image formation,preclassification,target alignment,and image classification.The results show that this method can improve the performance of image classification and speed up the data processing.