| With the rapid development of information technology,the field of Human-Computer Interaction has become the latest topic of current research,and user authentication is the first step towards Human-Computer Interaction.For decades,Personal Identification Numbers(PINs)and biometrics technology are still the most widely used authentication approaches.However,PINs are easily leaked to others,so it is particularly vulnerable to malicious attackers.In addition,biometrics usually requires specialized devices,which increases the cost.WiFi infrastructures are easy to deploy,widely used,and low-cost.Therefore,user authentication using WiFi signals has gained increasing attention,which aims to explore the mapping relationship between signal changes and human behavior.Therefore,this dissertation uses WiFi channel state information to extract keystroke behavior characteristics for user authentication.This dissertation implements user authentication based on WiFi channel state information from the perspectives of data collection,feature extraction,and classification recognition based on channel state information visualization,machine learning,and neural networks.The main work is as follows:1.Research on WiFi user authentication based on machine learning.Under the guidance of Fresnel theory and Rician distribution theory,we adjust the experimental setup and enhance the fingerprint information of keystroke in channel state.We collect data in three different scenarios,and the dataset contains 11(participants)× 200(input times)× 3(scenarios)=6600 data.Authentication was achieved by handcrafted extracting8 statistical features and using machine learning.Experimental results shows that the system achieves authentication accuracy is 85%,false accept rate is 12.8%,and false rejection rate is 11.2% among 11 participants.2.Research on WiFi identity authentication based on deep learning.We visualize the channel response as time-series images and use convolutional neural networks to learn the behavioral features in the energy and spectrum domains autonomously.These behavior features are used to realize user authentication.Empirical results show that the system achieves the average authentication accuracy is 90.5%,false accept rate is 7.5%,and false reject rate is 6% in the typical indoor environment. |
