Mutual information and clone selection algorithm based hyperspectral band selection method

Abstract

The large number of spectral bands in hyperspectral data provides abundant information to distinguish different land forms, however, the high dimensionality of hyperspectral data brings an extra computational burden. Hence band selection is important for hyperspectral data. The proposed hyperspectral clone selection algorithm based band selection method (CSABS) chooses an improved multi-dimensional mutual information method as the measure criterion, can select bands with richer information and lower redundancy from hyperspectral data and it is effective due to low reconstruction error. The clone selection algorithm is proposed as the search strategy where the adaptive clone and mutation operators are devised to speed up the convergence. This method not only take full advantage of multiple iterations of clone selection algorithm, but also has high accuracy. Experimental results on hyperspectral data demonstrate the effectiveness of the proposed band selection method.