基于人工神经网络的数字字符识别

提出一种用神经网络来识别含有噪声的数字字符的方法。神经网络采用带有动量项和自适应学习率的反向传播算法(BP)进行训练。样本由理想信号和带有噪声的信号组成。通过比较测试结果得出对同一网络既使用理想信号又使用带有噪声的信号对网络进行训练可使系统具有更强的容错性。最后给出的实验结果证明了该方法的有效性。

Number Character Recognition Based on Artifcial Neural Network

This paper uses artificial neural network to recognize print number characters.The system adopts back-propagation learning algorithm with momentum. The network model is composed of three layers with 35 input neurons, 10 middle neurons and 10 output neurons. It trains the network respectively using perfect signals and noised signals that represent the real world data. The method is capable of rapid classification, requires only fast, approximate normalization and preprocessing, and consistently exhibits better classification. It analyzes the results and discusses further questions to develop this model.