Research On Fingerprint Map Generation And Location Method Based On Generative Adversarial Transfer Leaning Networks Indoor Wi-Fi Location

With the wide coverage of Wi-Fi access points,Wi-Fi-based indoor positioning technology has become one of the hot research directions of indoor positioning technology.Existing researches are mostly aimed at Wi-Fi positioning in the same environment,establishing or optimizing positioning algorithms based on Wi-Fi location fingerprint database in multiple complex situations.Since the indoor communication of Wi-Fi signals is greatly affected by factors such as environment and time,the multiplexing rate of Wi-Fi location fingerprint data in different environments is low.This paper mainly focuses on the Wi-Fi positioning problem based on environment transfer learning,realizes the transfer learning of Wi-Fi signal knowledge in the source domain to the target domain,generates new Wi-Fi fingerprint map data in the target domain,and combines the control in the target domain to reduce the workload of the target domain data collection,and improve its positioning accuracy.The research work in this paper is mainly from the following two parts:1.Use Wi-Fi location fingerprint data from the source domain to generate the location fingerprint map of the target domain.Because the Wi-Fi signal propagation based on similar physical space has certain commonality,this paper analyzes the Wi-Fi signal propagation characteristics in different spaces to combat the depth transfer learning as the technical background,and designs the generative adversarial transfer leaning networks framework for Wi-Fi location fingerprint map,including generative adversarial networks,transfer learning networks,and supervisory information that unites the target and source domains.2.Use constraint transformation from the control point sample data of the target domain to transform,and achieve accurate location based on decision tree and Spearman distance.In order to further improve the positioning accuracy by the generated data,this paper mines features of the generated location fingerprint map,and takes the control points of the target domain as constraints,and then,the generated data are locally fitted with the control points as the center.Secondly,for the transformed data,this paper proposes the decision tree structure to carry out the coarse positioning of the grid by clustering of Wi-Fi signal vectors,and then calculates the signal feature metric based on Spearman distance.The signal vector of the point to be located is precisely matched with the signal vector in the cluster set to determine the actual physical space position.Compared with positioning algorithms such as SVM,KNN,and neural network,the positioning accuracy is higher.