Indoor Localization Algorithm Based On Transfer Learning And Fingerprint

With the rapid development of sensing,computation,wireless networking and communication technologies,many indoor location-based services have emerged,which promote technology of precise indoor localization in Internet of Things.Fingerprinting based localization has become attractive because of its open access and low cost.However,wireless signals are easily affected by environmental conditions,which leads to localization models built from previous radio maps inaccurate.Reconstructing radio maps and updating them with fewer labeled new data while maintaining high-accuracy positions is a key but difficult problem.Transfer learning addresses the lack of labeled data by attempting to learn information from one domain and transfer it to another related but different domain.Therefore,transfer learning is suitable to address the problem in fingerprinting based indoor localization where source and target domains have different distributions and the target domain retains limited labeled data.This thesis investigates the performance optimization of indoor localization algorithm based on transfer learning and fingerprint.The main research and innovation include:firstly,an indoor localization algorithm integrating fuzzy clustering with transfer learning is proposed.Fuzzy clustering is used to fit as much as possible the effects of environmental changes on a radio signal in an area.The results of fuzzy clustering are used to construct the transfer learning algorithm based on manifold alignment.The new radio map is reconstructed based on the results of transfer learning.Experimental results show that the proposed algorithm has lower location errors than that of those compared algorithms.Secondly,an indoor localization algorithm based on generalized kernelized linear discriminant analysis is proposed.Considering the advantage of linear discriminant analysis in data dimensionality reduction,the kernelized linear discriminant analysis algorithm is generalized as a transfer learning algorithm suitable for indoor localization,based on the characteristics of transfer learning and the application scenarios of indoor localization.Experimental results verify the effectiveness of the proposed algorithm.