改进卷积神经网络的验证码识别算法研究  

Research on Verification Code Recognition Algorithm Based on Improved Convolutional Neural Network

作　　者：高春庚[1] 罗俊丽[2] 路凯[2] GAO Chungeng;LUO Junli;LU Kai(School of Artificial Intelligence,Jiyuan Vocational and Technical College,Jiyuan 459000,China;College of Information Engineering,Xuchang University,Xuchang 461000,China)

机构地区：[1]济源职业技术学院人工智能学院,济源459000 [2]许昌学院信息工程学院,许昌461000

出　　处：《北京服装学院学报（自然科学版）》2023年第4期67-72,共6页Journal of Beijing Institute of Fashion Technology：Natural Science Edition

基　　金：国家自然科学基金资助项目(62101478);河南省科技攻关项目(212102210402);河南省高等学校重点科研项目(23B520016)。

摘　　要：在使用卷积神经网络(CNN)识别验证码的算法中,传统的算法需对卷积核参数进行随机初始化而存在网络收敛速度慢、训练时间长,识别准确率低的问题。提出一种基于主成分分析(PCA)和卷积神经网络相结合的验证码识别算法。首先,利用滴水算法对预处理后的验证码图片进行分割,然后对分割后的图片进行PCA学习,将学习得到的特征向量作为相应卷积层卷积核的初始化参数。实验结果表明,改进的方法在收敛速度和识别准确率方面均具有更好的性能。In the algorithm of using convolutional neural network to identify verification codes,the traditional algorithm randomly initializes the convolutional kernel parameters,which leads to slow convergence speed,long training time and low recognition accuracy.To solve this problem,this study proposed a verification code recognition method based on principal component analysis and convolutional neural network.The algorithm first used the drop-fall algorithm to segment the preprocessed verification code image,and then lused PCA learning for the segmented image,and took the learned feature vector as the initialization parameter of the convolution core of the corresponding convolution layer.The experimental results show that the improved method has better performance in convergence speed and recognition accuracy.

关 键 词：验证码识别 卷积神经网络 主成分分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]