Research On RFID Indoor Location Based On Neural Network

Indoor positioning has extensive application requirements in material management,security monitoring,automatic sorting,and other related industries.In addition,with the continuous improvement of intelligence in the above related industries,there is an urgent need to further improve the accuracy and intelligence of indoor positioning technology.Radio frequency identification(RFID)technology in indoor positioning technology has become a research hotspot due to its strong signal penetration,expand its function flexibly,low cost,popularization easily,strong anti-pollution ability,use sustainably,and other characteristics.The research content of indoor positioning based on RFID can be divided into relative positioning research and absolute positioning research.At present,the research on the above two positioning methods is not mature,and the typical problems include:the indoor relative positioning accuracy is not ideal in high-density tag scenes;in the case of limited reference tags,the error of indoor absolute positioning is relatively obvious.In view of the above problems,this paper proposes RFID indoor positioning methods based on neural network respectively on relative positioning and absolute positioning.The main works are as follow:(1)In order to improve the accuracy of relative location in indoor scenes with sequential detection requirements,this paper proposes a RFID relative location method based on a convolutional neural network(CNN).First of all,employing the idea of divide and conquer,we transform the problem of tags sequence detection into that of classification of the relationship between labels.Then obtain the left-right relationship between of tags through the convolutional neural network.Finally,to achieve the purpose of relative positioning,the tags are sorted based on error weights.Experimental results show that the accuracy of the proposed method can reach 90% and even more in the case of high density labels,namely the spacing of labels is 1cm.Compared with other methods,the accuracy of this method is greatly improved.(2)In order to reduce the error of absolute positioning in indoor scenes for obtaining two-dimensional position coordinates,this paper proposes a RFID absolute positioning method based on backward propagation neural network(BPNN).Firstly,the Gauss-Kalman filter is used to preprocess the data to reduce the interference of external factors.Then,the mean particle swarm optimization algorithm is improved by adaptive learning factors to reduce the probability of local convergence.After that,in order to avoid the negative impact of random initial value of the network,the improved particle swarm optimization algorithm is used to gain the optimal initial value of weights and thresholds of the network.Finally,the coordinate information of the target is predicted by the optimized network.Experimental results show that the proposed method can virtually control the positioning error within 30 cm.Compared with the traditional method,its positioning accuracy is improved to some extent.In addition,the corresponding prototype system is developed for the proposed methods,and is applied to the scenes of book sequence detection and simulated warehouse item location.