Research On The Adversarial Attack Algorithm To Enhance The Security Of The IoT Terminal Device

Today everything is interconnected,the number of human-computer interaction Io T terminal devices is increasing,while these devices bring convenience to humans,they also have many security risks.To solve the above-mentioned problems,this paper discusses the application of adversarial algorithm on Io T terminal devices and builds an algorithm verification platform,which is developed with a front and back-end separated architecture,consisting of an embedded artificial intelligence development kit,Docker,and MySQL.The following work has been accomplished in this paper.i.Research and Application of adversarial CAPTCHAsBased on the research and analysis of various types of CAPTCHA,the CAPTCHA images,which is highly secure and easy to combine with adversarial examples,is selected as the research object.A simple and efficient algorithm(PGD)and neural network target detection models are used to train adversarial CAPTCHAs.A database of massive CAPTCHAs was constructed for the CAPTCHA service center to call.ii.The design of algorithm verification interfaceBased on the research of current IoT terminal devices on the market,EAIDK was selected as the hardware development platform because of its rich interfaces and powerful performance.the graphical development framework PyQt5 was used to design the verification interface and the goal of effectively identifying the legitimacy of user input and preventing attackers is achieved.iii.The architectural design of validation centerDocker was used in the deployment of the validation center,enabling the migration of applications between different cloud platforms and improving the efficiency of subsequent development.When designing architecture,Considering that terminal devices may be attacked by DDo S,a high availability solution of Nginx proxy server combined with Keep Alived service and load balancing technology was proposed.In terms of database architecture,the MySQL master-slave architecture was used instead of the traditional monolithic to achieve database backup and read-write separation.iv.Test algorithm validation platformThe test programme of the platform was divided into two parts: the performance test of the adversarial CAPTCHAs and the function test of the validation center.In the performance test of the adversarial CAPTCHAs,m AP and adversarial perturbation were selected as evaluation indicators.The security performance of the adversarial CAPTCHAs under the application of three attack types was studied,then the impact of the adversarial CAPTCHAs on the degradation of the loss of model training was investigated,and finally the security performance of the adversarial CAPTCHAs in the face of the adversarial training model attack was studied.In the function test of the validation center,a back-end server test environment was built,the network connectivity and data transfer functions between client and server was tested.The results show that adversarial CAPTCHA technology can effectively identify the users,identify the validity of input,filter the traffic and improve the security performance of IoT terminal devices.