clustering is relatively few, which demands more attention in future work. In addition, deep learning has obtained remarkable achievements in the computer vision field, but it has few applications in the hyperspectral clustering arena. More effort and trials are needed in the future. Recently, graph-based clustering and subspace clustering have gained an increasing attention due to their relatively good clustering performance, and more and more algorithms have been proposed. (a) (b) (c) (d) (e) (f) (g) Corn-No-Till Soybeans-No-Till Corn-Minimum-Till Soybeans-Minimum-Till (h) Grass/Pasture Soybeans-Clean (i)(j) Grass/Trees Woods Hay-Windrowed Unlabeled FIGURE 13. Cluster maps of the different methods for the Indian Pines image. (a) The ground truth. (b) FCM. (c) FCM-S1. (d) CFSFDP. (e) GMM. (f) SC. (g) SGCNR. (h) FSCAG. (i) SSC. (j) L2-SSC. (a) (b) (c) (d) (e) (f) (g) Building 1 Grass-Synthetic (h) Building 2 Running Track Grass Bare Soil (i)(j) Trees Unlabeled FIGURE 14. Cluster maps of the different methods for the University of Houston image. (a) The ground truth. (b) FCM. (c) FCM-S1. (d) CFSFDP. (e) GMM. (f) SC. (g) SGCNR. (h) FSCAG. (i) SSC. (j) L2-SSC. 58 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE DECEMBER 2021