AN ENHANCED DENSITY PEAK-BASED CLUSTERING APPROACH FOR HYPERSPECTRAL BAND SELECTION

Recently, a fast density peak-based clustering algorithm, namely FDPC, has demonstrated its power on nonspherical clustering problems. In this paper, we propose an enhanced fast density peak-based clustering, namely E-FDPC, for hyperspectral band selection. The main contributions of the proposed E-FDPC, in comparison with the original FDPC are two folds. First, we introduce a parameter to control the weight between the normalized local density and intra-cluster distance. The other aspect is that, we present an exponential-based learning rule to adjust the cut-off threshold for different number of selected bands, where it is empirically defined in FDPC. Furthermore, an effective strategy, called isolatedpoint-stopping criterion, is developed to automatically determine the appropriate number of bands. That is, the clustering process will be stopped by the emergence of the isolated point (the only point in one cluster). Experimental results on real hyperspectral data demonstrate that E-FDPC approach could achieve higher overall classification accuracies than FDPC and other state-of-the-art band selection techniques.