论文部分内容阅读
本文受感知域划分思想的启发，将小波的多分辨分析与BP网结构相结合，构造了一种新的小波神经网络．该小波神经网络利用多分辨分析生成小波树，小波树的生长与网络的训练相结合，自适应地生成隐层结点，并且删除分类不佳的结点．以声纳信号进行了实验，结果表明：该网络充分发挥了小波的特点，将模式识别的特征抽取与分类器设计融为一体．
Inspired by the idea of perceptual domain division, this paper combines the multiresolution analysis of wavelet and BP network to construct a new wavelet neural network. The wavelet neural network uses multi-resolution analysis to generate wavelet tree. The combination of wavelet tree growth and network training can adaptively generate hidden layer nodes and delete badly-classified nodes. Experiments on sonar signals show that the network gives full play to the characteristics of wavelet and combines the feature extraction of pattern recognition with the classifier design.