Independent component analysis to hyperspectral image classification

Independent component analysis (ICA) is a popular approach to blind source separation. In this paper, we investigate its application to hyperspectral image classification. In particular, the performance of Joint Approximate Diagonalization of Eigenmatrices (JADE) algorithm is studied. The major advantage of using ICA is its capability of classifying objects with unknown spectral signatures in an unknown image scene, i.e., unsupervised classification. However, ICA suffers from computational expensiveness, which limits its application to high dimensional data analysis. In order to make it applicable to hyperspectral image classification, a data preprocessing procedure is employed to select the most important bands based on the band image quality. The number of bands ought to be selected is predetermined by an estimation method. The preliminary results from experiments demonstrate the potential of ICA in conjunction with band selection to unsupervised hyperspectral image classification.