A decision fusion approach for target classification
Xinhua Zhang, Liangji Lin, Jicheng Wang · 2002
This paper considers neural networks as an information engine and the fusion system based on these neural networks as an information network. First, the conditions to design an individual neural network model so as to enhance the performance of the combined classifier are proposed according to the information network theory. Next, the decision fusion is implemented by using fuzzy integral. In order to reduce the computation complexity and the conflict of existing evidences, a scheme selecting dynamically neural networks is presented. Finally, the proposed approach is applied to target classification of sonar system. Four neural network classifiers were obtained based on the designing conditions. Results showed that the classification accuracy and reliability of the fusion system were satisfactory.