基于神经网络的符号自动识别系统研究

提出了一种新的基于神经网络的目标符号自动识别系统。该系统在图像的二值化处理过程，采用了小波变换的方法，该方法可有效克服噪声的干扰，自动确定灰度图像二值化所需要的阈值。在符号识别部分，采用了双向联想记忆（BAM）人工神经网络技术，通过改进的感知器学习算法，增大了网络的容量，可实现对采集的有污染或缺损符号进行正确识别。仿真实验结果说明，系统具有较强的稳定性和有效性，且易于工程实现。

Automatic Recognition System for Target Symbol Based on Neural Network

Because traditional statistical decision methods and syntactic structural method have not good recognition rate under bad condition, it is necessary to use some advancing methods and ways to improve or propose some algorithm for desired recognition rate. This paper introduces a new automatic recognition system for the target symbol based on neural network. In the course of the binarization of image of the target symbol, a new way used in zero-crossing of wavelet transformation to select threshold is proposed. This way can overcome the effect of noise and select the threshold reasonably. In the course of symbol recognition, bidirectional associative memories (BAM) is used in this system. The system can efficiently recognize the polluted or damaged target symbol by using the learning method of developed perceptron. Experiment proves that stability and validity of the system are very strong, and it is easy to be implemented in the engineering.