基于改进Xception网络的验证码识别  

CAPTCHA Recognition Based on Improved Xception Network

作　　者：林开司[1,2] 张露 LIN Kaisi;ZHANG Lu(Department of Electrical Engineering,Tongling Vocational and Technical College,Tongling,Anhui 244000,China;School of Electronics and Information,Hangzhou Dianzi University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]铜陵职业技术学院电气工程系,安徽铜陵244000 [2]杭州电子科技大学电子信息学院,浙江杭州310018

出　　处：《福建技术师范学院学报》2024年第2期26-31,共6页JOURNAL OF FUJIAN POLYTECHNIC NORMAL UNIVERSITY

基　　金：安徽高等学校自然科学基金重点项目(2022AH052753).

摘　　要：验证码是一种公共自动化程序,用于区分用户和计算机.为了从网站大量获取信息,机器必须自动识别网站的验证码.为了自动识别验证码,研究基于深度学习的验证码识别,提出基于Xception网络和MLP的验证码识别方法.先利用Xception提取验证码特征,再经MLP标定不同权重,最终得到网络的最优权重分布.这种端到端的深度学习具有从输入到输出的预测,可以省去预处理、字符分割等步骤.经对不同验证码数据集的测试,该算法识别正确率在95%以上.Captcha is a commonly used automated program designed to differentiate between humans and computers.In order to collect large amount of information from websites,machines must be able to automatically recognize the website's captcha.In order to automatically recognize captcha,this paper conducts a study on deep learning-based captcha recognition and proposes a captcha recognition method based on Xception network and MLP.The Xception network is first employed to extract the captcha's features,then MLP is used to assign different weights and the optimal weight distribution of the network is eventually achieved.This end-to-end deep learning has the ability to predict the outcome from input to output,eliminating the need for preprocessing and character segmentation steps.Tests on different captcha datasets achieved a recognition accuracy rate of over 95%.

关 键 词：验证码 Xception网络 多层感知器 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]