Partial Least Squares Based Band Selection for Hyperspectral Imagery

Band selection for hyperspectral imagery is an efficient way to reduce its dimensionality.This paper proposes a new band selection method by introducing partial least squares(PLS) into the selection.First,the latent vectors are calculated with PLS.Then,the bands essential to the classification are determined according to the correlation between the bands and the latent vectors.The final band subset is obtained by analyzing the correlation between original bands in candidates.Experimental results show that,compared with other band selection methods,our method has better accuracy on classification.By avoiding features subset selection and eigenvalue decomposition of large matrix,our method performs much faster and more efficiently than traditional methods do.