Integration of Heterogeneous Classifiers Based on Choquet Fuzzy Integral

An information fusion method for heterogeneous classifiers based on choquet fuzzy integral is proposed in this paper. The output information transformation and the determining methods of fuzzy density which directly affects fuzzy measure are discussed in order to construct fuzzy integral fusion model. By simulating on sat image database in the benchmark real-world databases, it shows that the proposed method is able to classify the soil types and has better classification accuracy than individual classifiers. The comparison of the classification accuracy for the complex data with for the separated data of local-attributes is studied to demonstrate that better accuracy can be achieved for an appropriate sub-task than for a complex task. That is to indicate that the appropriate task segmentation will directly affect the accuracy of classification. The classification ability of Integration for heterogeneous classifiers is compared with of integration for same type classifiers. The results show that the former has better performance.