During aeromagnetic surveys using fixed-wing aircraft, magnetometers mounted inside the cabin are strongly affected by platform magnetic interferences. Existing aeromagnetic compensation models have difficulty in accurately modeling these complex magnetic interferences and often suffer from limited generalization capability. This paper proposes an intelligent compensation algorithm integrating a dual-fluxgate physical extension with an enhanced attention mechanism. First, an extended Tolles–Lawson (T-L) model leverages physical complementarity between dual fluxgates to enhance interference feature representation. Building on this, a physics-informed dual-branch parallel network architecture is designed. The physical branch dynamically models and decouples linear interference components, while the nonlinear branch introduces an improved attention mechanism to capture and remove non-stationary nonlinear interference in the signals. More importantly, the dual-branch network demonstrates superior generalization in level-flight extrapolation tests. Compared to traditional linear methods and pure data-driven models, the proposed approach reduces the residual standard deviation (STD) to 0.16 nT and achieves an improvement ratio (IR) of 22.31. This research significantly advances aeromagnetic compensation precision, offering a robust and high-performance solution for precision geophysical exploration.
Lei et al. (2026) studied this question.