Learning a Locality Preserving Subspace for ISAR Target Recognition

Io order to overcome the instability of Inverse synthetic aperture radar (ISAR) image caused by shift, rotation and scale variation, a new approach based on Locality Preserving Projections (LPP) algorithm of manifold learning is proposed to feature analysis in ISAR target recognition. Firstly, the LPP algorithm is used to reduce the dimensionality of the ISAR image, and then the reduced feature is classified by k-nearest neighbor classification with rejection recognition capability. Experimental results on four kinds of aircraft target suggest that the LPP algorithm has the capability of finding the low- dimensional manifold structure embedded in the high-dimensional ISAR image space, which is controlled by few parameters, such as attitude angle, scale and position, etc., and the better classification performance is acquired with the low-dimensional feature.