Research On Geomagnetic Indoor Positioning Algorithm Based On Deep Learning

With the rapid growth of indoor location service demand,indoor positioning schemes based on geomagnetic fingerprinting have attracted the attention of researchers due to their advantages of not requiring support from infrastructure equipment and having low costs.However,building a detailed geomagnetic fingerprint database requires a lot of time and manpower,and the recognition rate of the raw three-dimensional geomagnetic signal intensity as a geomagnetic fingerprint is relatively low.There are also information leakage issues in the positioning mode combined with the client and server.To address these issues,this thesis extends the geomagnetic fingerprint based on a generative model,designs a high-precision geomagnetic sequence fingerprint positioning algorithm,and builds an integrated positioning system consisting of a client and server.The specific research content and innovative points are as follows:1 Existing research methods generally have high manpower costs for building a detailed geomagnetic fingerprint database.In order to effectively reduce the workload of constructing the geomagnetic fingerprint baseline,this thesis proposes a novel conditional variational autoencoder generative model(CVAE-GAN)to construct a complete dataset.Based on the conditional variational autoencoder(CVAE),this model improves the adversarial idea in the Generative Adversarial Network(GAN).At the same time,the model uses the idea of the Deep Convolutional Generative Adversarial Network(DCGAN)to design the discriminator,using the Convolutional Neural Networks(CNN)architecture to replace the original fully connected network to improve the quality of generated fingerprint data.Experimental results show that by using the generative model,only 100 reference points need to be collected to achieve the same localization accuracy as 400 reference points collected without the model,reducing the workload by three times.2 To address the problem of low recognition rate of geomagnetic fingerprints,this thesis uses geomagnetic sequences to improve the identifiability of fingerprints.Based on the geomagnetic sequence fingerprint,this thesis proposes a Convolutional Neural Networks-Gated Recurrent Unit(CNN-GRU)geomagnetic sequence positioning algorithm based on the attention mechanism.This algorithm uses one-dimensional CNN and GRU to respectively extract the spatial and temporal features of the geomagnetic sequence fingerprint,and assigns different weights to the output of the GRU network at each time step through the attention mechanism to improve network performance.Finally,two optimizers are alternately used to train and optimize the network model.The experimental results show that the positioning effect of the positioning algorithm proposed in this thesis is significantly improved compared with the existing positioning algorithms in three indoor scenarios,and the highest positioning accuracy can reach 0.14 m.3 The existing positioning systems all adopt the client-server mode,which requires network communication and data transmission,and carries the risk of privacy leakage.To solve this problem,this thesis proposes an integrated positioning system for clients and servers and designs and implements a small,portable mobile device for location data collection.The mobile device uses Raspberry Pi(4B)as a data processor and storage device,and uses a nine-axis sensor to collect geomagnetic data.The overall size of the hardware is(10?8?8)cm~3.In addition,the indoor positioning algorithm of CNN-GRU based on the attention mechanism is implemented on this positioning system.