Hyperspectral image recognition based on KPCA

Hyperspectral image recognition is an important problem in practical hyperspectral imagery system. While nonlinear problem leads to identification problems, kernel method has provided a promising way to solve it. The performance of kernel-based algorithm is controlled by the appropriateness of kernel function and parameter greatly. However, simply adjusting the parameter of kernel is not effective enough because the data structures in kernel mapping space differ from each other when the parameters of kernel function differ. We present Kernel Principal Component Analysis (KPCA) applied on hyperspectral image. The learning system is improved by adjusting the parameters and kernel functions to the data structure for better effect on solving complex visual learning tasks. Experimental results proved the feasibility of the proposed methods.