Polarized Compass Orientation Framework Based on Real-Time Discrimination of Heading Information in Harsh Weather

How to achieve stable and reliable heading information in the case of harsh weather environments and global satellite navigation system (GNSS)-denied scenarios is more challenging for autonomous navigation. In recent years, the bioinspired polarized light orientation has been extensively and deeply explored in the field of autonomous navigation due to its high reliability and strong anti-interference performance. Herein, a novel polarized compass orientation framework in terms of real-time heading information discrimination is skillfully formulated and included in monitoring heading signal changes to characterize navigation behaviors without training complex neural networks or accumulating a large amount of data to drive. This study aims to solve the problem of the polarization compass orientation error caused by external environmental and internal device noise. Different from existing research, an optimized live heading signal discriminating and monitoring strategy based on sparse shrinkage local Fisher discriminant analysis (OSLFDA) is investigated in this article to distinguish the different data states comprised of pure heading signal and mixed noise without model traversal. Particularly, a discriminant function is calculated by formulating a heading data state discriminant matrix with the weight coefficient to maximize interclass and minimize intraclass separation. On this basis, an effective noise-suppressing algorithm based on a modified sliding singular spectrum analysis (MSSA), similarly without the need for a large amount of training neural networks, is presented to attenuate the noise in identified noisy components by OSLFDA so as to preserve the true heading signal to the maximum extent. Various representative experiment results demonstrate the efficacy and feasibility of the polarized compass orientation framework for drone navigation. In the drone-mounted test experiments on dust days, the root-mean-square error (RMSE, 0.5116°) is reduced by more than 81% when compared to the current up-to-date typical polarized light orientation techniques.