Neural network and ensemble method for hyperspectral image classification

Land cover classification is an important application of hyperspectral remote sensing. In this paper, we have proposed Neural Network and an ensemble method for automatic classification of selected hyperspectral data. At initial stage, K-means clustering is employed for unsupervised learning and knowledge discovery. Training data is prepared for neural network and tested with different scenes of spatial and spectral variations. The performance of proposed ensemble classifier is analyzed in terms of overall and average accuracy. Experiments are carried out on SAMSON hyperspectral data with 952 × 952 spatial dimension and 156 spectral bands. Four land cover classes are labeled for training and testing purpose. Experimental results demonstrate that neural network shows overall accuracy of more than 90% in almost all observations. Ensemble classifier is designed by employing support vector machine and neural network to further improve the accuracy up to 99%.