Discriminant independent component analysis for hyperspectral image classification

Abstract

A novel approach to discriminative extraction and independence of remote sensing hyperspectral imagery features using Discriminant Independent Component Analysis (DICA) is introduced in this paper. This approach maximizes the variance and negentropy of a given feature. The experimental results show that the proposed method has better classification accuracy when compared to using one of the most commonly used classifiers in the hyperspectral image of SVM. Classification accuracy obtained from DICA dimensionality reduction on Indian Pines dataset, value of average accuracy (AA) is 92.87 %, overall accuracy (OA) is 92.94 %, Kappa index is 0.82, whereas on Washington DC mall dataset, classification result obtained by using DICA dimensionality reduction is average accuracy amounted to 84.89 %, overall accuracy 84.69 % OA, Kappa index is 0.82. From experiment we can see that classification accuracy increases when using DICA approach compared than technique classical SVM.