BP算法的改进及用模拟电路实现的神经网络分类器

Improvement of Back-Propagation Algorithm and Neural Network Classifier Implemented with Analogue Circuits

基于用模拟电路实现神经网络分类器的目的，对多层静态前馈神经网络的ＢＰ算法做了改进，采用线性限幅函数代替Ｓｉｇｍｏｉｄ函数作为神经元的激活函数，给出了改进的ＢＰ算法。对该算法性能的实验研究表明：这种改进算法不但方便了用线性模拟集成运算放大电路实现神经网络，而且具有学习速度快，映射能力强等优点。根据本文算法设计的神经网络分类器，无论是计算机仿真，还是模拟电路实现，都得到了比较高的识别率。

or implementing neural network classifier with analogue circuits, an improvement of back-propagation (BP) algorithm of multilayer static feedforward neural networks is made. The linear threshold function is substituted for the Sigmoid function as an activation function of nerve cells. Therefore,an improved back-propagation algorithm is proposed. The performance of the improved algorithm is examined through experiments,it is shown that the improved algorithm not only can make it convenient to implement neural networks with linear integration operation amplifing circuits,but also have features of high speed in learning and strong mapping ability. According to the improved algorithm,a neural network classifier is designed. Both computer simulation and analogue circuits implementation are carried out with high discrimination.