A new methodology for determining fuzzy densities in the fusion model based on fuzzy integral

In the model of fusion of multiple classifiers based on fuzzy integral, the fused results are heavily dependent on the fuzzy measures which are defined on singletons and named fuzzy densities. Therefore, estimation of the densities or the measures is very important for the entire fusion process. Most of the existing methods regard the accuracy as an essential factor in constructing fuzzy densities. In this paper, the uncertainty of classifiers appeared during the classifying process is considered, and a new definition of fuzzy density which incorporated accuracy and uncertainty of the classifier is presented. A new method for determining fuzzy densities is proposed by considering both randomness and the cognitive uncertainty that is inherent in the source. This new method can reasonably measure the importance of each classifier and makes the performance of the fusion model improve significantly.