Indoor Location Method Based On Heterogeneous Transfer Learning

With the rapid development of mobile Internet,people's demand for location-based services based on location tracking is growing.The existing satellite positioning technology is only suitable for outdoor environment without obstacles,but it is difficult to provide accurate location information in complex indoor environment.Therefore,many scholars have proposed a large number of indoor positioning methods based on different sensors.However,there is a common problem in the existing indoor positioning methods,that is,the number and types of sensors will change with time,which makes the dimensions of samples in positioning and fingerprint database heterogeneous,resulting in the increase of positioning error.Transfer learning can learn knowledge from one domain and transfer to different but related domains.It has great advantages in dealing with the problems of distribution difference,dimension mismatch,insufficient label between source domain and target domain.In particular,heterogeneous transfer learning mainly studies the acquisition and transfer of knowledge from different domains in feature space,which can solve the heterogeneous problems in indoor positioning.Therefore,this thesis applies the idea of transfer learning to indoor location scene,and proposes two indoor location methods based on heterogeneous transfer learning.The work done in this thesis is as follows:1.Aiming at the heterogeneous problem that the number and type of sensors are partially missing,this thesis proposes a heterogeneous transfer learning method based on common features.The algorithm takes common features as the bridge,takes distribution alignment and divergence optimization as constraints,learns a mapping from common features to missing features,and fills the missing feature data in the target domain through this mapping,so as to solve the problem of location accuracy degradation caused by insufficient feature data.Finally,the experimental results show that the algorithm can achieve better location results when the target domain has complex feature data missing.2.In order to solve the heterogeneous problem that the number and types of sensors change dramatically,this thesis proposes a heterogeneous transfer learning method based on co-occurrence data.The algorithm takes co-occurrence data as a bridge,and projects the source domain into the feature space of the target domain by calculating a mapping,and transfers knowledge in the feature space.The algorithm has good adaptability to the change of environment without reconstructing fingerprint database.Finally,the experimental results in the measured data show that the algorithm can achieve better positioning accuracy under very few co-occurrence data.