ニューラルネットワークとファジィ集合を用いて拡張したDempster-Shafer理論によるパターン認識

概要

論文の詳細を見る
The authors propose a pattern recognition system, which uses several neural networks and extended Dempster-Shafer theory by fuzzy sets. Back-propagation networks are the most widely used of the neural network paradigms, and have been applied successfully in application studies in various fields. In most cases, the back-propagation networks have used as a pattern classifier. In the pattern classification, we need to deal with problems to classify correctly samples that are corrupted by noise. However, the single network is not ability to classify correctly the noisy patterns, even if their feature parts are presented in them. Firstly, we divide input vectors for training patterns into some small blocks to extract their feature parts, and make neural networks corresponding to the divided blocks of the input vectors learn their features. After training, each output value of the neural networks is considered as possibility degree expressed by membership functions in fuzzy sets theory. Therefrom, we introduce the output values expressed by fuzzy sets theoretically as a semi-mobile probability mass in extended Dempster-Shafer theory using fuzzy set theory. Then, the system classify the patterns from the outputs of the neural networks by extended Dempster-Shafer theory. Finally, to illustrate the effectiveness of the system, we apply it to pattern recognition of some capital letters with noise in typical example.
東海大学の論文