Research On Indoor Location Technology Of Location Fingerprint Based On Deep Learning

With the progress of technology and the popularity of mobile smart terminals,location related services have received more and more attention,which makes indoor positioning services popular research.Among many indoor positioning technologies,WIFI-based fingerprint positioning technology has become the main solution due to its simple equipment deployment,low cost,and easy signal access.However,due to the complex indoor environment,the signal propagation is affected by many obstacles,which leads to its low positioning accuracy in general.To address this problem,this paper applies deep learning technology to effectively improve localization accuracy.The main work of this paper is as follows:Firstly,the key factors affecting the performance of fingerprint localization are analyzed in terms of personnel flow,multipath effect,and data processing methods,and the experimental site is put up to collect fingerprint data required for localization,in the field it is separated at an interval of 1.6m,a total of 140 reference points is got.By repeatedly collecting 22 times in different periods on these reference points,about 20,000 fingerprints come into being.The data are also organized and pre-processed to establish the fingerprint database required for the study,and the path loss model validation showed that the dataset was reliable and usable.Secondly,a deep learning-based indoor localization scheme is designed.The scheme uses the RSSI image as input and uses CNN models to localize.The experimental results show that the localization accuracy of the indoor localization scheme based on the CNN model reaches 92.45%.To further verify the effectiveness of the scheme,the experiment is subsequently repeated using the publicly available dataset UJIIndoor Loc,and the results showed that the building localization accuracy was 99.14%,and the floor localization accuracy reached 81.27%.It is significantly better than KNN and SVM algorithms.Finally,considering that the CNN model cannot fully utilize the correlation characteristics between locations,this paper designs another scheme that introduces location correlation information by converting WIFI fingerprint data into sequence data and using the "memory" advantage of LSTM.The experimental results show that the localization accuracy of the LSTM-based model reaches 94.09% on the self-built fingerprint database,99.36% on the UJIIndoor Loc dataset for building localization,and83.74% for floor localization,which are significantly better than the CNN-based indoor localization schemes.