Abstract:Hyperspectral sensors produce very large images, with each pixel recorded at hundreds of different wavelengths. In this paper, a fast density peak clustering approach (FDPC), which was previously proposed for non-spherical clusters detection, is introduced to hyperspectral image classification. Since the centers generated by the unsupervised cluster algorithms may not always fit the expected classes, with which to be dealt, the supervised version of FDPC is presented. Moreover, spatial information are injected to enhance the local density. The proposed supervised spatial enhanced density peak classification approach (SSDP) addresses the aforementioned problem, especially for hyperspectral refine classification. Experimental results show that the proposed SSDP approach gives better performance.

A Supervised Density-Peaks-Based Classification Approach For Hyperspectral Images