摘要
手写体数字识别是模式识别的重要问题,反向传播模型(Back—Propagation)是一种性能较好的人工神经网络模型。本文采用与联想相结合的BP模型,对BP模型加以改进,并引入选举判决算法,既节省了训练时间,又提高了系统性能和识别率,并在ST—286H上用Turbo C2.0初步建立起一个带神经网络的手写体数字识别系统。
Handwritten numeral recognition is an important problem of pattern recognition. Although BP algorithm is good, its training time is too long, its dependence on training samples is too much, and its recognition rate is not satisfactory. BP model has been improved, threshold and feedback have been used. Association has also been taken in recognizing progress. Recognizing and training progress have been combined in training. In this way, the fault tolerance has been raised, the training time has been cut down and the dependence on the training samples has also been reduced. The Election Decision Algorithm has been Used in decision, it raises the character of the whole system. All these ideas have been realized on ST—286H computer with Turbo C2.0.
关键词
神经网络
手写体
数字识别
neural network, BP algorithm, association, election decision algorithm.
