To address non-stationary noise interference and azimuth boundary ambiguity of bionic polarization compasses in complex dynamic environments, research on an autonomous heading estimation algorithm is conducted. A Circular-Variational Bayesian Strong-Tracking Cubature Kalman Filter algorithm is proposed. In observation modeling, the von Mises distribution is introduced for directional statistical modeling, theoretically eliminating antipodal ambiguity caused by angle wrapping. Within the filtering framework, Variational Bayesian (VB) inference and Strong Tracking (ST) theory are deeply coupled: the former is utilized to adaptively estimate time-varying observation noise covariance online to suppress environmental interference; the latter introduces fading factors to maintain residual orthogonality, enhancing transient responsiveness during sudden environmental changes and carrier maneuvers. Experimental results demonstrate that the algorithm achieves seamless fusion of multi-modal bionic cues, stabilizing the heading estimation error within 0.6° under strong interference scenarios. This architecture overcomes the limitations of traditional filtering in handling periodic boundaries and non-stationary noise, providing a robust computational framework for highly reliable bionic navigation.