摘要
由于目标运动及其所处环境的复杂性,雷达目标数据之间往往呈现出局部的非线性,如果采用传统的线性子空间方法降维,必将会使雷达目标识别性能有所下降,基于以上原因,文章尝试将流形学习的思想应用于逆合成孔径雷达(ISAR,inverse synthetic aperture radar)目标二维像的目标识别。局部保持投影(LPP,locality preserving projections)是一类有效的流形学习算法,但它在构建权矩阵时没有充分利用样本的类别信息。针对此问题,提出了一种称为局部保持判别投影(LPDP,locality preserving discriminant projections)的子空间学习方法,该方法通过构建类内和类间两个权矩阵来描述多类样本数据集的局部几何结构,以使在高维空间中相互靠近的同类数据点在低维嵌入空间中也相互靠近,而不同类的近邻点则尽可能地远离。对三类飞机目标的仿真实验结果表明,与PCA、LDA和LPP等算法相比,LPDP算法具有更好的识别性能。
It is well known that the relationship between different radar targets is often nonlinear due to the complexity of target＇s movement and environments, so the recognition rate will decrease when the traditional linear dimensionality reduction methods are used. For this reason, the idea of manifold learning is introduced to Inverse Synthetic Aperture Radar （ISAR） 2D image target recognition. Locality Preserving Projections （LPP） is an effective method of manifold learning, but it does not make full use of the class label information when the adjacency matrix is constructed. In this paper, we propose a new subspace learning method called Locality Preserving Discriminant Projections （LPDP）. In LPDP, the local geometry of multi-class samples are described by the within-class adjacency matrix and the between-class adjacency matrix, and the data points are mapped into a subspace in which the nearby points with the same label are close to each other while the nearby points with different labels are far apart. The simulated experimental results about three aircraft targets indicate that the LPDP algorithm has better classification performance than those of PCA, LDA and LPP.
出处
《中国电子科学研究院学报》
2009年第6期656-660,共5页
Journal of China Academy of Electronics and Information Technology
关键词
流形学习
局部保持投影
ISAR二维像
目标识别
manifold learning
locality preserving projections
ISAR Image
target recognition
