RFID And Deep Learning Based Indoor Localization

With the popularity of Internet of Things(Io T)devices,wireless signalbased precise positioning technology has developed rapidly in recent years and has spawned a large number of intelligent applications in the industry.Important applications include product sequencing on conveyors and object positioning and inventory on shelves.These applications require precise positioning of target objects.As one of the important supporting technologies of the Internet of Things,radio frequency identification(RFID)technology has been widely used in indoor positioning of smart factories or warehouses.At present,RFID-based indoor positioning technology has made some progresses,and its accuracy has reached the centimeter level under ideal conditions.However,due to the environmental noise and multipath effects in real application scenarios,the existing positioning models will suffer severe performance loss.Only a few existing studies have considered this problem and proposed corresponding technical improvements,but these improvements still have large limitations or require additional hardware equipment.To address such problem,this study uses a small amount of prior data,combined with the traditional RFID hologram synthetic aperture model and deep learning algorithms to propose a Deep Learning Enhanced Holography(DLEH)technology,which can be generate an adaptive RFID-based localization model(ARLM)for a stable environment.The generated model can accurately locate RFID tags even in environments with strong multipath interference.The contributions of this paper include:(1)A more robust synthetic aperture algorithm for RFID holograms is proposed.The existing research on RFID synthetic aperture holograms considers the noise in the definition of the likelihood function and proposes some methods to reduce their effects,but the performance is still not good in the environment of multipath effect interference.In this study,after observing the effects of multipath effects on RFID signals and the likelihood distribution of RFID holograms,this study design joint hologram and adjacent hologram,which can provide a more robust likelihood value for the target tag.(2)A position estimation algorithm based on RFID hologram is designed.Existing RFID hologram models use the position corresponding to the maximum likelihood value to estimate the tag position.However,the ubiquitous environmental noise makes this method that relies on a single likelihood value not robust.This paper treats holograms as images,and uses deep learning algorithms to analyze the likelihood distribution in holograms,and accurately estimate the spatial position of target labels.(3)An adaptive RFID positioning model training method is designed.As deep learning algorithms are introduced into RFID positioning,this paper needs to design efficient training methods.In order to take advantage of both experimental data and simulation data,a hybrid training method is designed in combination with transfer learning technology,which greatly reduces training costs while ensuring model positioning accuracy.This paper uses commercial RFID equipment to implement a deep learning enhanced hologram technology prototype,and collects a large amount of data in the real scene to evaluate the performance of the adaptive RFID positioning model.The experimental results show that the RFID adaptive positioning model generated by DLEH can obtain centimeter-level accuracy in a two-dimensional plane using a single antenna even in a multipath environment.