一种基于改进核主成分分析的SAR图像识别方法研究

SAR Images Recognition Based on Modified Kernel Principal Component Analysis

摘要:

针对传统核主成分分析方法识别SAR图像时,存在图像像素之间关联性差、对目标姿态角依赖性强等局限性,研究了一种基于改进核主成分分析的SAR图像识别方法。其研究思想是,结合SAR图像的特点提出了一种基于局部特征核主成分分析的特征提取方法,并设计了一种基于灰关联分析的双分类器对提取特征进行分类。MSTAR仿真实验表明:该方法不仅可以增强图像像素之间的相关性,而且对目标姿态角不存在依赖性,仿真结果验证了方法的有效性和可行性。

Abstract:

The SAR image recognition method based on the traditional kernel principal component analysis exist some problems, such as weak correlations between the pixel and strong dependence of target azimuth. In this paper, a new SAR image recognition method based on local feature kernel principal component analysis was proposed to overcome these problems, and a double classifier based on gray correlation analysis was also presented. Experimental results with MSTAR dataset show that not only the correlations between the pixels can be strengthened, but also the dependence of target azimuth is disappeared, and this method is effective and feasible.