改进的BP神经网络在数字识别上的应用

Application of improved BP neural network to digit recognition

首先介绍了传统的人工神经网络方法对数字字符的识别,进而在变换函数、误差函数以及惯量项等方面对学习算法进行了改进,提出局部自适应算法——RPROP算法,使网络具有一定的容错能力,用VC完成对数字字符识别的模拟。最后实验表明,改进的算法可以有效地完成对训练样本的识别,并且弥补传统方法学习速度低、平均误差大的缺点。

The artificial neural network method used to recognize number character is introduced. The learning algorithrn is improved in the case of transition function, error function, inertia and so on by the local adaptive algorithm- RPROP. Therefore the network has certain fault-tolerance capability and can use VC to simulate the number character recognition. The experiments show that the methods can recognize the number efficiently and make up the disadvantage of the traditional methods that cannot learn rapidly and has large average errors.