为了更好地对合成孔径雷达(SAR)图像进行分类,文中提出一种基于局部线性嵌入方法(LLE)和随机拉普拉斯特征映射方法(SLEM)相融合的算法。在拉普拉斯特征映射方法的基础上引入随机过程的概念,并将局部线性嵌入方法与随机拉普拉斯映射方法进行函数融合,两种方法的融合为提取高维空间中嵌入的低维特征提供了更详细的结构信息,保留了原始数据集的几何特征;将算法应用于MSTAR数据集,再通过KNN分类器进行分类;最后,实验结果证明了该算法的有效性。

Synthetic aperture radar(SAR) image has high resolution and large amount of information. In order to better classify SAR image, this paper proposes an algorithm based on the fusion of locally linear embedding(LLE) and stochastic Laplacian eigenmaps(SLEM). The concept of random process is introduced based on the Laplacian Eigenmaps, and the LLE is combined with the SLEM is used for function fusion. The fusion of the two methods provides more detailed structural information for extracting low-dimensional features embedded in high-dimensional space and retains the geometric features of the original data set. The algorithm is applied to MSTAR data and then classified by KNN classifier. The experimental results show the effectiveness of the algorithm.

TN957. 32

国家自然科学基金资助项目(61774137);山西省自然科学基金资助项目( 201801D121026,201701D121012,201701D221121);山西省回国留学人员科研资助项目(2016-088);中北大学2017 年校科研基金资助项目(2017027)