字符识别中改进的神经网络算法设计

如何提高字符识别的速度和准确率在车牌识别系统中是很关键的问题。传统的BP算法可以实现非线性函数的映射，经过有监督式的学习规则可以达到比较好的识别效果。但是BP算法识别速度慢，而且容易陷入局部最小。引入动量因子可以平滑误差曲面梯度方向剧烈变化的作用，从而在一定程度上解决了局部最小值的问题。仿真结果表明了该算法的有效性。

The design of developed neural networks arithmetic in identifying the character

How to improve the identifying speed and accuracy in the character identifying system of the number plate is a key problem.The nonlinear function map can be realized with the classical BP arithmetic and the identifying result will be relatively good under the study rule of supervising type.But BP arithmetic has a low identifying speed and easy to encounter local minimization.In this article,the momentum factor introduced can smooth sharp change of the grads orientation,therefore,the local optimization was solved in some degree.The simulation result demonstrated the improved arithmetic's effectiveness.