摘要
在本文中,我们用统计模式识别的方法分析了目前在模式识别中得到广泛应用的多层BP神经网络,揭示了具有线性输出单元的多层BP神经网络用作特征提取器和分类器时具有良好性能的原因。同时,我们设计了一个使用多层BP网络作为特征提取器和分类器的、普适的工件识别系统。对不能识别的样本,采用模糊推理技术,把传统的直观特征识别结果和多层BP网络结果在特征级上融合,提高系统的性能。
In this paper, we analyze the widely used Multi-level BP Neural Network (MBPNN) with traditional statistics pattern recognition method and reveal why MBPNN with linear out-units has very well performance when used as feature extraction and pattern recognition. On the same time, we design a generalized part recognition system with MBPNN. As to the samples that have not been able to be recognized we adopt fuzzy logic to fusion obvious feature and MBPNN's feature, and increase the performance of the system.
出处
《计算机应用与软件》
CSCD
1997年第3期9-17,34,共10页
Computer Applications and Software
关键词
神经网络
模糊推理
模式识别
BP网络
Neural network, fuzzy logic, pattern recognition, membership func- tion, BP algorithm.
