基于多尺度稀疏字典的SAR图像目标识别方法

针对合成孔径雷达目标识别问题,提出一种基于多尺度稀疏字典的SAR图像目标识别方法。稀疏字典选择是稀疏表示中的关键问题之一,该方法利用小波多尺度分析构造过完备稀疏字典,将训练样本图像在小波解析域中进行小波多层分解,充分利用小波多尺度分析突出图像局部特征的特点,并和过完备稀疏表示有效结合组成级联字典。通过求解测试样本相应的稀疏系数矢量并根据系数矢量中对应训练样本类别的重构误差判定目标类型。实验结果表明,该方法在识别前无需对SAR图像进行预处理,具有良好的识别效果。

SAR ATR Based on Multi-scale Sparse Dictionary

A new approach is developed for Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) based on multi-scale sparse dictionary. The construction of the dictionary is a crucial issue in SAR ATR under the framework of sparse representation. The wavelet multi-scale analysis is used to construct the sparse dictionary so that local characteristics can be better studied. The training images are decomposed by using wavelet multi-scale analysis in wavelet domain,and the sparse coding for characteristics of each scale is represented by using multi-scale sparse dictionary. The class that the testing sample belonged to is determined by the minimum reconstruction error from the sparse parameter vectors under the framework of the cascade dictionary. The effectiveness of the method is proved by the experimental results.