| A NEW MULTIPLE CLASSIFIER SYSTEM BASED ON A PSO ALGORITHM FOR THE CLASSIFICATION OF HYPERSPECTRAL IMAGES |
| Paper ID : 1082-SMPR |
| Authors: |
|
Sayyed Hamed Alizadeh Moghaddam *1, Mehdi Mokhtarzade2, Sayyed Ahmad Alizadeh Moghaddam2 1Faculty of Geodesy and Geomatics Engineering, K. N. Toosi University of
Technology, Tehran, Iran. 2Faculty of Geodesy and Geomatics Engineering, K. N. Toosi University of Technology, Tehran, Iran. |
| Abstract: |
| Multiple classifier systems (MCSs) have shown great performance for the classification of hyperspectral images. The requirements for a successful MCS are 1) diversity between ensembles and 2) good classification accuracy of each ensemble. In this paper, we develop a new MCS method based on a particle swarm optimization (PSO) algorithm. Firstly, in each ensemble of the proposed method, called PSO-MCS, PSO identifies a subset of the spectral bands with a high J_2 value, which is a measure of class-separability. Then, an SVM classifier is used to classify the input image, applying the selected features in each ensemble. Finally, the classification results of the entire ensembles are integrated using a majority voting strategy. Having the benefit of the PSO algorithm, PSO-MCS selects appropriate features. In addition, due to the fact that different features are selected in different runs of PSO, diversity between the ensembles is provided. Experimental results on an AVIRIS Indian Pine image show the superiority of the proposed method over its competitor, named random feature selection method. |
| Keywords: |
| Multiple classifier systems (MSCs), Particle swarm optimization (PSO), Hyperspectral image classification |
| Status : Conditional Accept (Poster) |
