Research And Application Of Captcha And Handwriting Based On CNN

Convolutional neural network is a hot topic in the research and application of deep learning in the field of AI image recognition.The optimization of its model has important application value in various character image recognition.Verification code is designed for system server-side identification,which is operated by robot computer or human user.It plays an important role in network system,APP application login,account registration,information release,password retrieval and other verification links.With the development of AI research,the risk of automatic identification of verification code by computer algorithm is higher and higher The optimization and improvement of technology and captcha generation algorithm can help the system avoid automatic attack or brute force cracking by malicious programs.Handwritten numeral recognition belongs to the research field of intelligent character recognition.At present,handwritten numeral recognition has not been widely applied to the education industry,such as automatic recognition of test paper evaluation scores and output to the score sheet,digitization of students' historical status information files,etc.The practical value of unified intelligent engineering can reduce the workload of teaching staff.The main research work of this paper is as follows:Firstly,this paper studies the theory and technology of graphic captcha.Based on the investigation and summary of convolutional neural network character recognition algorithms and models at home and abroad,this paper studies and builds a deep learning framework tensorflow as the platform,based on the alexnet model,and takes two kinds of problems of Jizhong vocational college educational Administration campus network system login captcha and teachers' handwriting marking recognition as the research objects,Complete the theory,technology and data preparation of convolutional neural network graphic verification code and handwriting recognition.Secondly,a text captcha image recognition model(Alexnet + TL + BN)based on transfer learning and batch normalization is proposed,designed and implemented.The convolution core size and network structure of alexnet classic network model are optimized and improved.Batch normalization is introduced into some volume layer and full connection layer.By using migration learning,the improved alexnet model is pre trained by using the generated verification code data set and MNIST handwritten data set.A large number of experiments are carried out on the test set of acquisition and processing.The analysis results of the experimental data show that the recognition rate of the deep learning algorithm is very high,and the verification code of the research object system has the potential security risks of violent cracking or automatic registration,Then,the improvement method and the corresponding strategy of verification code are proposed for different vulnerabilities of the same kind.Thirdly,the improved convolution neural network model is used to recognize the score of handwritten test paper.The mainstream MNIST data set and the handwritten score of test paper collected on the spot are used as training samples to test and recognize the data set,recognize the student number,score and other handwritten character information of candidates,and provide technical support for the design and development of application system.Fourth,the design,development,testing and preliminary application of the campus character recognition application system are carried out.The requirements analysis,UI and function requirements analysis and design,overall design,logic design and detailed design are carried out.The system implementation and system testing are carried out by using pyside2 development system GUI and python platform architecture.The scores of the test paper are identified and automatically output as the score sheet,which improves the efficiency of the college's educational administration in entering student status records.The content of this paper has positive and negative effects on Intelligent Security:mutual game,which can be used to crack and prevent.The research work has a certain reference value in increasing the security of campus network system and further application of intelligent educational administration.