An Adaptive FCM Probabilistic Neural Networks with Confidence Criteria

Abstract

A novel PNN classifier for underwater vehicle noise source recognition is proposed. Such PNN classifier based on adaptive FCM algorithm and confidence criteria. Confidence criteria recognition technique allows the classifier recognize the abrupt noise without any abrupt noise samples to train base classifier, which is difficult to the traditional PNN classifier. The adaptive FCM algorithm can optimize classifier topology structure and save recognition time. Experimental results show adaptive FCM-PNN which has better generalization performance and real time performance than RBF neural network and the traditional PNN, and can recognize the abrupt noise effectually through confidence criteria.