摘要
设计出一种用于识别具有旋转和变形特点的字符型验证码的卷积神经网络,分别采用小类和大类字符集的方法训练神经网络,再使用不同的学习速率和样本数量进行训练。通过实验分析,得出在使用大类神经网络,选择适当的学习参数,并且增大样本数量的情况下,可以有效提高识别率的结论。
A convolution neural network were designed for recognizing the CAPTCHA that had a rotation and deformation characteristics. Subcategories and categories character set were used to train the neural network. It was followed by the use of different learning rate and number of samples for training. By analyzing the experimental results obtains in the case of using the categories of neural network, the appropriate learning rate was selected. Increasing the number of samples could effectively improve the recognition rate.
出处
《莆田学院学报》
2016年第2期63-66,共4页
Journal of putian University
基金
福建省教育厅科技项目(JB13190)
福建江夏学院校级教育教学改革项目(J2014T002)
关键词
卷积神经网络
验证码
旋转
变形
convolutional neural network
CAPTCHA
rotation
deform