基于加权纹理特征的SAR图像目标识别算法

SAR image target recognition algorithms based on weighted texture features

由于SAR图像与光学图像的显著差异,光学图像的目标识别算法并不能应用到SAR图像中,因此研究SAR图像的目标识别具有重要的意义。传统的基于模型的SAR图像目标识别算法中将所有的特征同等看待,然而不同特征对于目标分类的贡献度可能差别很大。给不同的特征赋予不同的权重,可能会改变目标在特征向量空间中的相对位置,从而给出更合理的识别结果。采用SAR图像的纹理特征作为分类特征,在支持向量机分类算法中加入使用ReliefF算法计算得到的特征权重。试验结果表明这种加权后的目标识别算法具有更高的目标识别率。

Target recognition algorithms of optics images can＇t be applied to SAR images because of the prominent difference between SAR images and optics images, which makes the research to SAR image target recognition more important. Traditional recognition algorithms based on models pay same attention to all features by giving the same weight, whereas different features may have quite different influence to the recognition results. The relative positions of target projected to feature space may change by giving different features with different weights, which helps to get more appropriate recognition results. This article uses texture features as the recognition features, then the SVM algorithms is adopted with the feature weights calculated by ReliefF algorithm. The test results show that this weighted recognition algorithm has higher recognition rate.