A New Band Selection Algorithm for Hyperspectral Data Based on Fractal Dimension

Feature selection especially band selection plays important roles in hyperspectral remote sensed image processing. It is worth nothing that band selection approaches need to be combined with image spatial structure information so as to select valid bands and improve the performance. But all of the existing remote sensing data processing algorithms are used for the conventional broadband spectral data and can not process high dimensionality data effectively and accurately. According to the characteristic of HRS data, the algorithm which named optimal band index (OBI) based on fractal dimension was put forward in this paper. In OBI algorithm, firstly, the fractal dimension was used as the criterion to prune the bands which have noises, and the bands which have better spatial structure, quality and spectral feature were reserved. After that, the correlation coefficients and covariance among all bands were used to compute optimal band index, and then the optimum bands were selected. At last, in the experiment the proposed algorithm was compared with the other two algorithms (Adaptive Band Selection and Band Index), it proves that the OBI algorithm can work better on the band selection in hyperspectral remote sensing data processing than other algorithms. * Corresponding author. Email: [email protected], [email protected], Tel: +86-13851706937