A New Hyperspectral Image Classification Method Based on Extended Wavelets Transform

Abstract

In the past few years, more machine learning frameworks have been applied to hyperspectral image classification tasks and they have achieved good results. Most classification methods often ignore the correlation between local spatial features. They treat each pixel vector independently. In this paper, a new hyperspectral image classification method is proposed in which both spatial and spectral information is caried out by using wavelets transform. In this method, 1D wavelets transform is applied to the spectral dimension of the HSI to reduce spectral dimensionnality. Then, 2D wavelets transform is used to extract the edge texture and spatial information of the hyperspectral image. Finally, spectral and spatial features are fused to classify the images using support vector machine (SVM) classifier. Experiments are carried out on the Indian Pines dataset and the obtained results show the effectiveness of our proposed approach compared with conventional approaches for HSI classification.