An Adaptive Denoising Algorithm for RTD Fluxgate Sensor time differential Signals Based on Isolation Forest and VMD-Wavelet Fusion

The residence time-difference (RTD) fluxgate sensor is critical for high-precision magnetic field measurements in applications such as unmanned aerial vehicles (UAV) magnetic surveying, geological exploration, and environmental monitoring. However, measurement accuracy and system stability are often compromised by environmental and internal noise. To address this, this paper proposes a multi-stage fusion processing algorithm combining anomaly detection, modal decomposition, and entropy weight fusion. The method first detects anomalies using dynamic thresholding based on the Isolation Forest (IForest) model and removes outliers via a weighted neighborhood mean compensation strategy. Variational Mode Decomposition (VMD) optimizes modal parameters dynamically, constructing an entropy-weighted multi-scale signal decomposition framework. A hierarchical adaptive wavelet threshold function is then applied to reconstruct and suppress noise in the modal components. Experimental results demonstrate significant improvements: compared to IForest-wavelet, wavelet transform, and local-attention transformer denoising, the signal-to-noise ratio (SNR) increases by 39%, 13%, and 107%, respectively, while the mean absolute error (MAE) decreases by 4%, 3%, and 14%. This algorithm effectively enhances signal accuracy and stability, providing a robust, adaptive solution for high-precision denoising of RTD fluxgate sensors. It offers critical technical support for precise magnetic measurements in UAV exploration, deep geological resource localization, and environmental pollution monitoring.