Integrating Feature Optimization with Truncated Total Least Squares for Aeromagnetic OBE Interference Compensation

• A systematic aeromagnetic compensation framework is proposed, integrating dual-threshold feature selection, EMD-based feature anchoring reconstruction, and TTLS to mitigate complex electromagnetic interference. • The proposed method achieves a superior compensation precision of 7.94 nT, outperforming traditional linear models and existing advanced methods by over 60.96% and 25.86%, respectively, across diverse flight scenarios. • The method enhances multi-scale profiling of electromagnetic interference, while simultaneously eliminating systemic biases induced by bilateral noise. • Demonstrates significant potential for Earth remote sensing missions, such as geophysical exploration and GNSS-denied autonomous navigation. Aeromagnetic survey is a crucial technique for geomagnetic field measurement. To address the degradation of measurement accuracy caused by severe on-board electronics (OBE) interference, this paper proposes a systematic aeromagnetic compensation method. The method is designed to mitigate common issues in existing research, including feature redundancy, inadequate analysis and utilization of feature components, and noise in input variables. First, at the feature selection level, a method based on dual-threshold greedy iteration is proposed. By synergistically optimizing the correlation between features and interference alongside the independence among features, a subset of electrical features with high interpretability and low redundancy is effectively extracted. Subsequently, at the feature correction level, electrical features are deconstructed at multiple scales using Empirical Mode Decomposition (EMD). A weighted reconstruction algorithm incorporating anchor constraints is developed to adaptively enhance high-contribution components, thereby significantly mitigating long-term drift in aeromagnetic data. Finally, at the compensation modeling level, Truncated Total Least Squares (TTLS) is employed for regression modeling. By incorporating bilateral noise from both inputs and outputs into a unified modeling framework, the method effectively suppresses systematic biases induced by input noise. Experimental results show that the proposed method achieves an average compensation accuracy of 7.94 nT across four independent test sets under various flight conditions, improving upon traditional linear models by 60.96% and existing advanced methods by 25.86%, thereby providing high-precision data support for applications such as geomagnetic navigation and mineral exploration.