Research On The WiFi Sensing Recognition Method Based On Integrated Learning

In recent years,WiFi sensing and recognition technology based on machine learning has attracted wide attention.However,the existing solutions have two main drawbacks.One is that when the number of sensing target categories increases,the accuracy decreases seriously.Secondly,due to the fact that machine learning models generally use probability measures to make decisions,which are constrained by the forced probability distribution,the sensing model is less stable under the actual conditions.This thesis mainly focuses on the above two problems,and puts forward the corresponding solutions.Furthermore,the method is evaluated on two existing sensing applications(gait recognition and gesture recognition).Experimental results show that the method proposed in this thesis can achieve high precision and stable recognition of large-scale sensing objects categories.The specific research contents are as follows:(1)In order to solve the problem of the decline in the accuracy of wireless sensing models in large-scale categories,this thesis proposes a method of sensing model selection based on integrated learning.This method integrates multiple sensing-based models by fusing multifeature and multi-classification algorithms.When recognizing test samples,the model selection mechanism is used to determine the optimal sensing-based model for recognition,so as to overcoming the defect of low accuracy of a single sensing model under large-scale categories.(2)In order to solve the instability of the current sensing model based on probability measurement,this thesis proposes a sample filtering method based on conformal prediction model.The basic idea is to apply statistical measures to make classification decisions,overcome the constraints of probability measures,and establish a sample filtering mechanism to filter test samples.In practical application,filter the false recognition samples caused by the instability of the sensing model,and further enhance the stability of the sensing model through incremental learning of the model.(3)Based on the above two methods,a WiFi sensing recognition system based on integrated learning is designed and implemented,and a comprehensive and detailed evaluation of the system is carried out on a large number of WiFi datasets.The experimental results show that with the continuous increase of the number of sensing object categories,the average recognition accuracy of the system in gait recognition and gesture recognition increases from 56.3% and 51.8% to 85.33% and 92.6% respectively,and provides consistent and stable performance.In addition,the filtering method of the system can achieve an error filtering rate of less than 5%.After incremental learning of the model,the system performance can maintain stability.