Semisupervised hyperspectral image classification with SVM and PSO

Abstract

This paper proposes a novel semisupervised approach to classify hyperspectral image. This method can overcome the limited training samples problem. It combines support vector machine (SVM) and particle swarm optimization (PSO). The new approach exploits the wealth of unlabeled samples for improving the classification accuracy. The method can inflate the original training samples by estimating the labels of the unlabeled samples. The label estimation process is performed by the designed PSO. The effectiveness of the proposed system is carried on a real hyperspectral data set. The experimental results indicate that the classification performance generated by the proposed algorithm is generally competitive. © 2010 IEEE.