Dimension reduction for hyperspectral image based on the second generation bandelet transform

Abstract

A dimensionality reduction method is proposed by using the second generation Bandelet transform. The redundant components of the hyperspectral cube are firstly partitioned into several subsets. Subsequently the Bandelet coefficients and the geometries flows of the hyperspectral image are generated by performing second generation Bandelet transform.In the follow step, Principal Components Analysis (PCA) is introduced to simplify the redundant data. Finally, the new reduced hyperspectral cube is reconstructed by taking inverse Bandelet transform. Some numerical simulations are made to test the validity and capability of the proposed dimensionality reduction algorithm.