基于非线性流形学习的ISAR目标识别研究  

ISAR Target Recognition Based on Non-linear Manifold Learning

作　　者：何强[1] 蔡洪[1] 韩壮志[1] 尚朝轩[1] 

机构地区：[1]军械工程学院光学与电子工程系,河北石家庄050003

出　　处：《电子学报》2010年第3期585-590,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60572062)

摘　　要：详细分析了逆合成孔径雷达(Inverse Synthetic Aperture Radar,ISAR)二维像的非线性流形结构特点,指出ISAR二维像可以看作是由位置、姿态和尺度等内在参数共同作用而张成的一个在高维图像空间中的非线性流形.在此基础上,论文将非线性流形学习的思想引入到ISAR目标识别领域,提出了一种基于局部保持投影(Locality Preserving Projections,LPP)算法和k近邻分类器的ISAR二维像特征提取和目标识别方法.该方法首先利用LPP算法对维数较高的ISAR二维像进行降维,然后采用具有拒识功能的k近邻分类器对四类飞机目标进行了分类识别.仿真实验结果表明,LPP算法能够发现嵌入在高维ISAR图像空间中的低维非线性流形,并且能够利用LPP算法降维后的特征获得较高的识别率.The non-linear manifold structure property of inverse synthetic aperture radar（ISAR）images is analysed intensively,and it is pointed that the ISAR images can be viewed as a non-linear manifold of high-dimensional ISAR image space controlled by a few parameters,such as position,attitude and scale.The idea of non-linear manifold learning is introduced into ISAR target recognition,a new feature extraction and recognition method for 2-D ISAR images based on Locality Preserving Projections（LPP）algorithm and k-nearest neighbor classification is proposed.Firstly,the LPP algorithm is used to reduce the dimensionality of the ISAR images,and then four kinds of aircraft target are classified by k-nearest neighbor classification with rejection capability in the low-dimensional subspace.The simulated experiment results suggest that the LPP algorithm has the capability of finding the low-dimensional manifold structure embedded in the high-dimensional ISAR image space,and a higher recognition rate is acquired with the low-dimensional feature obtained by LPP.

关 键 词：目标识别 ISAR二维像 非线性流形 局部保持投影 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN951[自动化与计算机技术—计算机科学与技术]