Ensemble clustering model of hyperspectral image segmentation

Abstract

In this paper, the method of ensemble clustering model of hyperspectral image segmentation is proposed. We select several abundant information and identifiable band from each hyperspectral face images cube using Principal Component Analysis, in order to alleviate the computation burden and improve the clustering performance. K-means base clustering is performed on the all selected bands respectively, and different initial clustering center values were given to each band. This solves the K-means over-reliance on the initial clustering center values. Finally, we use the automatic integration method based on factor graph to fuse the results of base clustering and gain the more robust cluster result.