Research On Monitoring Method Of Learning State Based On WIFI Channel State Information

In recent years,under the wave of the development of the Internet of Everything,the technology of the Internet of Things has developed rapidly and continuously innovated.Among them,wireless perception has become one of the core technologies in the field of human action recognition,using wireless signals to identify some actions in daily life,such as fall detection,human behavior detection and sleep monitoring,has great application value.Nowadays,students spend more time studying at home,their learning efficiency is difficult to guarantee,and they are easily disturbed by things around them without parental supervision.At this time,a method that can assist parents to monitor students’ learning conditions is needed.In view of the above requirements,this paper proposes a learning state monitoring research method based on WiFi channel state information(CSI)and using convolutional neural network(CNN): Wi-LSM.This scheme realizes data collection by forming AP mode of mini industrial computer equipped with Intel 5300 wireless network card and Linux802.11 n CSI Tool driver,through research and analysis of the influence of human body characteristics on the wireless signal propagation process,then carry out the corresponding data processing and feature extraction,use a deep learning network to further train the extracted signals to find deeper features,and establish the relationship between the wireless signal and the action category,so as to realize the monitoring of the learning state.First,the collected raw data containing a large amount of information needs to be processed,and the information obtained by the appropriate antenna is selected as the sample feature.In view of the phenomenon of packet loss in the data taken by using commercial WiFi,in order to ensure its integrity,it is necessary to fill the missing data packets in the CSI conjugate matrix through linear interpolation.The outliers are eliminated by the Hampel filter,and then the high-frequency noise is denoised by the Butterworth filter function,which preserves the original features of the data to the greatest extent.Second,in view of the waste of phase information in the past,unwrapping,linear transformation and denoising are performed in the phase aspect to eliminate the original phase confusion and offset problems.In this paper,the amplitude and phase information are integrated as the input of the neural network feature.In terms of amplitude,in order to obtain rich time-frequency information,wavelet transform(WT)is used to further extract the extracted time-domain information to obtain the frequency-domain information of higher-level clear signals and phase difference information are combined to form a two-dimensional matrix as the original feature of the input network training,which is more likely to obtain more useful information,thereby effectively reducing the number of The computational complexity greatly improves efficiency while reducing noise.Third,the convolutional neural network model optimized by adding channel attention(SENET)multi-layer mechanism combines time-frequency domain and phase difference joint information for action recognition,which improves the accuracy and robustness of the system.The comparison and verification show that the best detection rate of the experiment can reach 96% in the indoor environment,which has a good classification effect.Moreover,the learning state monitoring research method based on WiFi channel state information proposed here has less cost,easy collection and strong adaptability,which solves the inconvenience of using machine vision for learning state monitoring in the past.Identification has certain research significance.