Hyperspectral Image Classification with Support Vector Machines Based on the Maximum Noise Fraction

Abstract

In this paper, a Hyperspectral Image (HSI) classification algorithm is developed based on the Support Vector Machine (SVM) and the Edge-Preserving Filtering (EPF). To avoid dimensional disaster in classification algorithm, Maximum Noise Fraction (MNF) transformation is performed before SVM. It not only reduces the computational cost, but also reduces the effect of noise and improves the classification accuracy. To realize MNF dimension reduction, multiple linear regression is utilized to estimate the noise covariance matrix in MNF. Experimental results show that both the accuracy and computational efficiency of HSI classification are improved by the proposed algorithm. Moreover, it performs robust to noise.