Abstract:A parallel processing model based on multi-scale decomposition is proposed to improve the polarized light orientation accuracy caused by harsh external environments, different types of internal device noise, and complex attitude changes of carriers during actual motion. The maximum similar eigenvalue of the autocorrelation matrix, which can effectively reflect the complexity of the detected sequence signal, is proposed to classify the band-limited intrinsic mode functions decomposed based on the central frequency variational mode decomposition method into low to high frequency components. By this way, accurate noise separation from the original noisy heading data is achieved when noise and useful signals are mixed. Multi-scale time-frequency peak filtering with the advantage of long and short windows is adopted to adaptively remove noise in low frequency components, and a gated recurrent unit neural network is employed to model and compensate the heading data error in high frequency components synchronously. The turntable dynamic experimental results show that the orientation accuracy is 0.2373°on sunny days, which is 85.08% higher than the existing research methods. The UAV dynamic orientation accuracy is 0.2037°, which is 89.59% higher than existing research methods the same as turntable experiment. The effectiveness and practicality of the proposed method can meet the requirements of strong autonomy and strong anti-interference ability of polarized light orientation under modern and future electronic countermeasures conditions.