Innovative Magnetic Localization for Close-Range Scenarios: Bi-GRU With Adaptive Sensor Selection and Spatial Feature Extraction

Indoor magnetic target localization can employ distributed sensor arrays, however, the accuracy in the vicinity of the sensors is often limited by two primary factors: the distance requirements of magnetic dipole modeling and the occurrence of outliers from individual sensors. To overcome the limitations, this article introduces a bidirectional gated recurrent unit (Bi-GRU)-enhanced close-range magnetic localization algorithm. The proposed method can adaptively filter out abnormal sensor data, enhancing close-range positioning performance under a distributed sensor array. Specifically, the algorithm employs a threshold-based strategy to dynamically identify and remove outliers, enabling the optimal selection of sensors. Additionally, a coordinate rotation technique is proposed to address data imbalance issues during localization. To enhance the precision of localization, a Bi-GRU-based spatial feature extraction method is introduced, capable of handling nontime-series data with missing values. Adaptive weighted fusion is incorporated to further improve localization accuracy. Simulation results demonstrate that the proposed method achieves a positioning error as low as 0.0413 m within a spatial range of $2{\times }2{\times }2$ m, outperforming other existing methods under the same conditions, particularly in close-range positioning scenarios. Furthermore, experimental validation on a real-world magnetic field dataset confirms that the proposed method attains a positioning accuracy of up to 0.72 cm within a distance range of 4.8–28 cm.