Classification of hyperspectral images based on progressive bands selection and eigenmaps techniques
Résumé
Hyperspectral images provide richness information of scene. However, the classification of these images remains a challenge due to the huge amount of data volume, the redundant of information and the high correlation between the bands. The dimensionality reduction has been developed in the literature to solve these issues. Generally it's divided into two main categories: projection techniques and bands selection techniques. In this paper, we propose an adaptive dimension reduction approach for classification of hyperspectral imagery based on progressive bands selection and eigenmpas technique in order to obtain an accurate classification. In fact, we improve the eigenmaps technique by incorporate the clusters of relevant bands obtained with progressive bands selection in order to take advantage to measure the amount information presents in each band. The performance of the proposed approach were evaluated using AVIRIS hyperspectral images and the obtained classification with SVM classifier shows the effectiveness compared to the obtained classification without applying the dimension reduction of original hyperpsectral image.