Hyperspectral Image Recognition Using SVM Combined Deep Learning

Abstract

In this paper we present the conbination of deep learning and Support Vector Machine applied on the recognition of hyperspectal images. Hyperspectral image recognition is an essential problem in the practical hyperspectral imagery system. While deep learning is capable of reproducing feature vectors with great dimensions out of original data, it leads to great time cost and the Hugh phoenomenon. Such nonlinear problem is regarded as obstacles and kernel method appears to be a promising way to solve it. The performance of kernel-based learning system is influenced by the choices of kernel function and parameter greatly. We present the kernel learning method termed Support Vector Machine (SVM) applied on feature vectors supplied by deep learning upon hyperspectral image. The learning system is improved by adjusting the parameters and kernel functions to the data structure for improving performance on solving complex tasks. Experimental results validate the feasibility of the proposed methods.