Fuzzy Recognition Based on the Total Matching Degree of Multi-features
Lei Jingtao, Zhu Jianmin · 2011
This paper presented a novel modular recognition method, which was the Fuzzy Neural Network Inference Recognition (FNNIR) method based on the total matching degree computing of multi-features. The Modular Recognition Functional Component (MRFC) was composed of three software modules, which were feature extraction module, off-line learning module and on-line fuzzy inference module. A zero-order Takagi-Sugeno Fuzzy Neural Network was constructed to obtain non-linear mapping between multi-features and actual classification. The bell-shaped membership function was proposed to describe the distribution of feature values. The parameters of the membership function for each feature can be determined by off-line learning of FNN. The matching degree of each feature and the total matching degree can be calculated by the membership functions, and the recognition result can be determined by the total matching degree. The experiment results show that the average test error of FNN is only 0.005443, with high modeling accuracy, thus making it suitable for on-line applications with high recognition accuracy.