深度学习下网站验证码的破解与防范
|
摘要
随着互联网的快速发展,越来越多的网站为了防止用户利用机器人自动注册、登录、灌水,都采用了验证码自动识别技术。然而,简易的验证码依然存在安全风险,为了防止网络爬虫或者恶意攻击的登录操作,本文将通过PaddlePaddle平台进行验证码识别的深度学习,阐述深度学习下网站验证码的破解和防范。实验结果表明:将训练实例的验证码数据训练成预测模型,使用预测模型来自动识别验证码中包含的内容实现验证码自动识别,而将验证码的字符位置随机摆放,可有效防止攻击者对图片进行裁剪分析。
With the rapid development of the Internet, more and more websites have adopted captcha automatic identification technology in order to prevent users from using robots to automatically register, login and chatter. However, simple captchas still have security risks. In order to prevent the login operation of web crawlers or malicious attacks, this paper will carry out the deep learning of captchas identification of PaddlePaddle platform and explain the cracking and prevention of captchas under deep learning.The experimental results show that the captcha data onto the training example is trained as a prediction model, and the content contained in the captcha is automatically identified by the prediction model, while the random placement of the characters of the captcha can effectively prevent the attacker from cutting and analyzing the image.
With the rapid development of the Internet, more and more websites have adopted captcha automatic identification technology in order to prevent users from using robots to automatically register, login and chatter. However, simple captchas still have security risks. In order to prevent the login operation of web crawlers or malicious attacks, this paper will carry out the deep learning of captchas identification of PaddlePaddle platform and explain the cracking and prevention of captchas under deep learning.The experimental results show that the captcha data onto the training example is trained as a prediction model, and the content contained in the captcha is automatically identified by the prediction model, while the random placement of the characters of the captcha can effectively prevent the attacker from cutting and analyzing the image.
引文
[1]何福泉,李伟烽,林培娜,等.验证码的识别技术分析与研究[J].甘肃科技纵横,2019,48(2):1-4.
[2]基于卷积神经网络的12306验证码识别[D].华南理工大学,2017.
[3]AICon 2018开幕百度PaddlePaddle引开发者热议[J].智能机器人,2018(1):19.
[4]尹宝才,王文通,王立春.深度学习研究综述[J].北京工业大学学报,2015(1):48-59.
[5]张铮,王顺帆,董雷.基于深度学习的验证码识别[J].湖北工业大学学报,2018,33(2):5-8+25.
[6]陈少镇,王培彬,蓝汝琪,曾梓鑫,杨森.移动互联网+师生互动的课程表设计与应用[J].电脑知识与技术,2018,14(04):76-79.
[7]刘欢,邵蔚元,郭跃飞.卷积神经网络在验证码识别上的应用与研究[J].计算机工程与应用,2016,52(18):1-7.
[8]杨航,蔡浩.中文图片验证码识别方法的研究[J].汕头大学学报(自然科学版),2018,33(3):47-53.
[2]基于卷积神经网络的12306验证码识别[D].华南理工大学,2017.
[3]AICon 2018开幕百度PaddlePaddle引开发者热议[J].智能机器人,2018(1):19.
[4]尹宝才,王文通,王立春.深度学习研究综述[J].北京工业大学学报,2015(1):48-59.
[5]张铮,王顺帆,董雷.基于深度学习的验证码识别[J].湖北工业大学学报,2018,33(2):5-8+25.
[6]陈少镇,王培彬,蓝汝琪,曾梓鑫,杨森.移动互联网+师生互动的课程表设计与应用[J].电脑知识与技术,2018,14(04):76-79.
[7]刘欢,邵蔚元,郭跃飞.卷积神经网络在验证码识别上的应用与研究[J].计算机工程与应用,2016,52(18):1-7.
[8]杨航,蔡浩.中文图片验证码识别方法的研究[J].汕头大学学报(自然科学版),2018,33(3):47-53.