基于稀疏流形学习的SAR图像识别

SAR Image Recognition Based on Sparse Manifold Learning

稀疏性是SAR图像的一个显著特征,而且SAR图像存储模式的维数很高,要对其进行识别存在很多困难.为了解决上述问题,提出一种基于稀疏流形学习的SAR图像识别方法.首先进行图像增强,以保持目标的边缘结构信息;其次利用样本协方差矩阵的谱范数确定能得出数据低维流形的最少数据点数;再利用拉普拉斯特征值映射(LE)的核化方法计算样本外点的低维嵌入;最后采用KNR分类器进行识别.MSTAR仿真实验证明了其可行性,并与其它识别方法进行比较,验证了其优越性.

Sparsity is a remarkable character of Synthetic Aperture Radar（SAR） image and its dimension of storage is high,so the recognition of SAR image is very difficult.In order to solve the problem,an algorithm of SAR image recognition based on sparse manifold learning is proposed.Firstly the image was enhanced in order to preserve the edge information of the objective;The second step was determining the least number of points which can get the integrate low-dimensional manifold by the spectrum of the sample covariance matrix;then utilized kernel extending of Laplacian Embedding（LE） to get the low-dimensional coordinates of the out-of-sample,at last SAR images were recognized by Kernel-based Nonlinear Representor（KNR）.Experimental results on MSTAR show its feasibility and superiority by comparing with other methods.