Classification of Hyperspectral Image with Feature Selection and Parameter Estimation

This paper presents a new method for hyperspectral image classification. It combines support vector machine (SVM), particle swarm optimization (PSO), and genetic algorithm (GA) together. Its aim is to improve the classification accuracy and reduce the computation consumption based on heuristic algorithms. Because the classification accuracy is impacted by the parameters of the SVM model and feature space in the training and testing steps. In order to optimize the parameters and feature subset, our proposed technique integrates the GA operators into PSO. The effectiveness of the proposed system is carried out on a real hyperspectral data set. Comparison for the classification performance with other reference classifiers is also reported. The obtained results clearly confirm the superiority of the proposed hybrid algorithm.