Hyperspectral Band Selection Based on Affinity Propagation Clustering

Band selection can preserve the physical meaning of hyperspectral data while reducing dimension, and has application in many aspects. The cluster of affinity propagation (AP) algorithm is according to the correlation of data points, and the AP algorithm regards all data points as potential clustering centers. We propose a band selection method based on AP clustering, which uses spectral information divergence and spectral correlation angle (SID-SCA), and spectral information divergence and spectral gradient angle (SID-SGA) to improve the similarity calculation in AP algorithm. The reducing dimension results arc input into the support vector machine (SVM) classifier to classify, and the classification accuracy is calculated and verified using the data set Indiana Pines. The experimental results reveal that the proposed method can better extract the information of the band and obtain a high classification accuracy.