NavFormer：运动坐标系下的IGRF预测

Triaxial magnetometer components fluctuate with sensor attitude, even when the International Geomagnetic Reference Field (IGRF) total intensity target remains invariant. NavFormer is designed to forecast this invariant target by leveraging rotation-invariant scalar features and incorporating a Canonical Symmetric Positive Definite (SPD) module. This module plays a crucial role in stabilizing the spectrum of window-level second moments of the magnetometer triads, effectively mitigating sign discontinuities. Specifically, the Canonical SPD module constructs a canonical frame for each data window using a Gram matrix. Within this canonical frame, it applies state-dependent spectral scaling, addressing the complexities introduced by attitude variations in magnetometer data within moving coordinate systems. This approach ensures accurate prediction of the invariant IGRF total intensity. Through this innovative feature extraction and spectral stabilization technique, NavFormer enhances the robustness and precision of geomagnetic navigation and positioning in dynamic environments. It offers a novel solution for applications requiring high-accuracy geomagnetic field information, such as inertial navigation systems and UAV attitude control. The model reduces strong reliance on attitude estimation prevalent in traditional methods, directly extracting stable features related to the total geomagnetic field intensity from raw magnetometer data, thereby simplifying algorithmic complexity and improving real-time performance.