Recognition On Internet Banking CAPTCHAs With Convolutional Neural Network

Internet technology is developing at a rapid speed.While the network brings great convenience to people's life and work,it also generates various network security risks,such as registration of a large number of junk accounts,insecurity of web login,and hacker hacking.Confidential data,etc.,in order to prevent these problems from happening,the CAPTCHA can effectively prevent the robot's free activities,its operation is simple and can guarantee the security of the network,and is favored by the landing pages of various websites.At present,most online banking CAPTCHAs use text-type CAPTCHAs to protect the security of bank websites.Therefore,verifying the security of online banking CAPTCHAs is of great significance to the network security of banks.At present,most websites use text CAPTCHAs to protect the security of websites,and online banking is no exception.The text CAPTCHA is small in size and can be loaded quickly,so its popularity is high,and there are many people who naturally study its security.The text CAPTCHA gradually evolved from simple characters to the current replicating,distorted,sticky,and non-fixed CAPTCHAs,along with the constant updating and development of the cracking algorithm.This paper mainly uses the online banking CAPTCHAs as the research object,and proposes a CAPTCHA identification method based on convolutional neural network for CAPTCHAs identification problem.For the gradation of the CAPTCHA image with adhesion and rotation,it is used as a whole input.By constructing a multi-layer convolutional neural network and selecting the Relu function as a nonlinear excitation function,the character extraction and CAPTCHAs of the CAPTCHAs image are performed.Identification.The Tensorflow deep learning framework is used to train the convolutional neural network model.Based on the LeNet-5 model,the appropriate convolutional neural network structure is constructed.The deep learning platform Floydhub is used to carry out the CAPTCHA model training,and the model is evaluated through the test set.Parameter tuning.The test results show that the model can effectively identify CAPTCHAs with different complexity and has good robustness and generalization ability.To some extent,the problem of identification of the hard-to-segment CAPTCHA is solved.Through the process of cracking the CAPTCHA,the inadequacies of the online banking CAPTCHAs are found,so that some suggestions for improving the CAPTCHAs are proposed for protecting the security of the banking network.