Automatic Fuzzy Clustering Based on Adaptive Multi-Objective Differential Evolution for Remote Sensing Imagery

Traditional automatic fuzzy clustering methods can obtain the optimal number of clusters by maximizing or minimizing one single-objective function using validity indexes. However, the effectiveness of these methods depends on the selection of the validity indexes, and one single-objective function may not provide satisfactory results because of the complexity of remote sensing images. For instance, the same land types may have different spectral curves, and different land types can have similar curves. To avoid this problem, this paper proposes a novel automatic fuzzy clustering method based on adaptive multi-objective differential evolution (AFCMDE) for remote sensing imagery. In AFCMDE, the automatic clustering problem is transformed into a multi-objective problem using two objective functions: J_m and the Xie-Beni index. AFCMDE is designed as a two-layer system comprising an optimization layer and a classification layer. In the optimization layer, AFCMDE searches for a feasible number of clusters by minimizing the J_m value and the Xie-Beni index. Based on the obtained number of clusters, AFCMDE utilizes non-dominated and crowd-distance sorting to obtain the optimal clustering centers and output the clustering results. In addition, a self-adaptive strategy without user-defined parameters is also used to improve the differential evolution. Experimental results using three different types of remote sensing image show that the AFCMDE algorithm consistently outperforms the other traditional clustering algorithms and the previous single-objective automatic fuzzy clustering algorithms.