Research And Implementation Of Captcha Recognition Based On Deep Learning

With the development of information technology,information technology has brought people a new perspective to understand the world,allowing people to obtain information more quickly,and bringing great convenience to people's work and life.But it brings convenience to people and also brings some security risks of network information.The captcha acts as a gatekeeper of the web portal,and as the first security line of many website systems,it is more and more widely used in the network.Captchas are widely used in account registration,system login,information confirmation,etc.,in order to prevent some malicious automated programs from attacking the network.At present,the most widely used captchas are mainly text-captchas,image-captchas,and voice-captchas.Text captchas are the most widely used type of captchas.The main reason is that text captchas are low in cost and easy to apply.Therefore,text captchas are used as an example in this article.The main work of this article is reflected in the following three aspects:1.In order to solve the problem that the captcha data is difficult to collect,this paper,by writing program,automatically generates the captcha suitable for the data set,mainly generates the pair of captcha data set for the Generative Adversarial Network training and can generate the verification with different types of characters and different character length according to the requirements.2.Design a scheme that can remove the interference information in the captcha through conditional generation of the confrontation network,reduce the technical application in the pre-processing process of the captcha,and improve the work efficiency.Due to the characteristics of the generation of the confrontation network,the universality of the model is realized.3.The end-to-end identification of the captcha is realized by constructing a convolutional neural network with inception structure.This method can effectively reduce the loss of information by the captcha during the segmentation stage,and can improve the recognition rate of the captcha.In addition,the use of Inception in the network can improve the speed of network training and recognition.4.A captcha generation and recognition system is implemented using B/S architecture.The system can not only generate experimental data,but also identify captcha.