Study On RFID Indoor Positioning Algorithms Based On Deep Learning And Location Fingerprint

As a non-contact wireless information technology,radio frequency identification technology(RFID)has excellent performance in the areas of article management,smart logistics,smart medical systems,and personnel tracking.How to effectively use the RFID positioning system has good research value and broad application prospects for locating indoor targets and obtaining the information of the target.At present,a series of algorithms have been proposed for RFID positioning systems,For example,based on time-of-arrival(TOA),time-difference-of-arrival(TDOA),angle-of-arrival(AOA),received-signal-strength-indicator(RSSI)location algorithms,and LANDMARC Scene Sensing algorithm.Although they improve positioning performance to varying degrees,they are only suitable for simple indoor scenarios.However,compared to outdoor positioning,the indoor positioning faces multi-path propagation,shadow effects,non-line-of-sight and other complex electromagnetic transmission environments.In addition,with the increasing application of RFID positioning systems,the positioning range will continue to increase,and the positioning environment will become increasingly complex.Traditional range-based RFID indoor positioning algorithms are difficult to solve the above problems.This paper first studies several common scene-aware RFID positioning algorithms: LANDMARC method,nearest neighbor method,K-nearest neighbor algorithm,weighted K-nearest neighbor algorithm,and naive Bayes method,and analyzes and compares them through simulation.Based on this,aiming at the existing problems in traditional RFID indoor positioning algorithms,the algorithms based on deep learning theory and position fingerprinting for RFID indoor positioning are further studied.The work is supported by the project “RFID multi-label three-dimensional positioning methods based on position fingerprinting and deeplearning in complex scenes”(No.20180101329JC),which is supported by the Jilin Provincial Natural Science Foundation of China.The creative works of this thesis are as follows:With response to the problem of poor positioning accuracy and long positioning time caused by the increasing positioning environment and the increasingly complex positioning environment,the deep neural network(DNN)and position fingerprinting algorithm are combined,and an RFID indoor positioning algorithm is proposed based on DNN and position fingerprinting.The path loss model is used to model the signal propagation in the room and obtain the received signal strength to form a fingerprint database.The DNN is used as a regression model to find the relationship between the fingerprint data and the position information to achieve accurate real-time positioning of the position fingerprint.The algorithm has higher positioning accuracy and shorter positioning time in the presence of path loss and noise,which is suitable for locating multiple targets in a large indoor location environment.To further reduce the positioning time required by the positioning system and improve data processing capabilities and fitting capabilities,the deep belief network(DBN)and position fingerprinting algorithm are combined.An RFID indoor positioning algorithm and its improved version are proposed based on DBN and position fingerprinting.The improved algorithm chooses to use the unsupervised learning characteristics of the DBN to extract features of the fingerprint database.The obtained features are compared with the characteristics of the fingerprint data to be tagged after learning through the DBN,so as to accurately locate the positioning tags.Using MATLAB and Python simulation analysis,we can see that both algorithms have higher positioning accuracy than traditional algorithms.