Research And Implementation Of Indoor Magnetic Positioning Technology Based On Deep Learning

Geomagnetic fingerprint indoor positioning technology based on deep learning can realize accurate positioning services without infrastructure support,which has attracted the attention and research of many experts and scholars.However,most of the existing indoor localization methods use raw three-dimensional geomagnetic signals as the localization source,and the geomagnetic fingerprints constructed by such methods are less distinguishable,especially in the large-scale indoor environment with a similar structure.Meanwhile,the built-in magnetometer of the smart terminal is easily affected by factors such as the user's posture and walking speed.This feature will not guarantee the reliability of positioning services provided by indoor positioning applications.In addition,it is difficult for deep learning models deployed on cloud servers to provide instant response services,and it is necessary to further reduce the response time of indoor positioning applications based on deep learning models.In view of the above problems,the main research work of this paper can be divided into the following three points:·A method for constructing magnetic sequence fingerprints based on the sliding window is proposed.The key to indoor positioning technology based on geomagnetic fingerprints is the identifiability of geomagnetic fingerprints.This paper expands the characteristics of geomagnetic data and proposes a neighbor average interpolation algorithm and a method of removing redundant data at an equal interval to solve the problem of different lengths of sequence data.And on this basis,a sliding window-based geomagnetic fingerprint construction method is designed to improve the identifiability of geomagnetic fingerprints.·An indoor localization method based on CNN-LSTM model is proposed.Combining the advantages of one-dimensional convolutional neural network and long short-term memory network in processing sequence data,a CNN-LSTM hybrid neural network model is designed and constructed to extract magnetic sequence fingerprint features,and multiple optimizers are introduced to train and optimize the CNN-LSTM model for improving the positioning performance of indoor localization methods.·An indoor positioning application framework under cloud-edge-end collaborative computing is designed and implemented.Giving full play to the complementary and synergistic relationship between cloud computing and edge computing,an indoor positioning application framework under cloud-edge-end collaborative computing is designed and implemented.The application framework realizes the lightweight deployment of indoor positioning services by establishing an indoor positioning application task model,designing a computing offloading strategy,and building a cloud-edge-end collaborative computing model.Based on three typical indoor environments(laboratory,science building corridor,student dormitory corridor),this paper explores and evaluates the he positioning performance of the localization method in this paper,and compares it with the existing localization methods.The experimental results show that the localization performance of the CNN-LSTM localization model proposed in this paper is not only better than the existing indoor localization methods based on deep learning models,but also achieves good positioning performance when using different brands of smartphones and walking with different speeds.In addition,the indoor positioning application under the cloudedge-end collaborative computing mode can effectively reduce the response time of location service requests,and further improve the reliability of indoor positioning applications by maintaining and managing the localization model.Therefore,it can achieve the effective support for indoor positioning application scenarios with high real-time and reliability requirements.