A Study On The Method Of Multi-classifiers Fusion Based On Fuzzy Integrals

Multi-classifiers fusion is a powerful solution to the difficult pattern recognition problem. Fuzzy Integral is an aggregation tool in multi-classifier fusion, which canimprove the accuracy of classification and the robustness of systems. Inmulti-classifier fusion based on fuzzy integrals, fuzzy measures have much influence on the performance of fusion systems. If the fuzzy measures are well defined, theaccuracy of classification can be improved distinctly. However, if the fuzzymeasures are badly defined, it is possible that most classifier's accuracy of classification is higher than the fusion system's accuracy. In this thesis, three types of uncertainties about the method of Multi-classifiers Fusion based on Fuzzy Integrals studied: methods to confirm the fuzzy measure, accuracy of the classification and practicability. The optimizations of the fuzzy measure respectively based on the PSO and the prey-predator are particularly discussed in detail.Firstly, the common classifiers are introduced and analyzed. From the conclusion, the strongpoints and weaknesses of themselves are discovered, and the multi-classifiers fusion is educed. Then this thesis introduces the common means of the multi-classifiers fusion.Secondly, the model of Multi-classifiers Fusion based on Fuzzy Integrals studied, and the common methods on confirming the fuzzy measure are particularly discussed in detail. On the conclusion, the shortage of the common methods is discovered.Finally, contraposing the insufficiency of the single-classifier and multi-classifiers fusion, this thesis give the new metods to confirme the optimum fuzzy measure, which use the PSO and the prey-predator arithmetic respectively. The methods are verified to be practical by experiments, and they can enhance the accuracy of classification evidently and the efficiency of classification.