Residual Estimation of the Angle of Polarization Based on K-nearest Neighbors Regression in Differential Polarization Navigation System

Bio-inspired polarization navigation has gained much attention in recent years due to its powerful autonomy and non-cumulative errors. But when it is applied to marine ships, the polarization information obtained directly from the skylight will inevitably be disturbed. Noisy conditions can be because of the harsh atmosphere in the marine atmospheric boundary layer (MABL). Real-time estimation and compensation of such non-occluded types of atmospheric interference are an important means to improve the accuracy of polarization orientation. In this paper, a residual estimation method of the angle of polarization (AoP) based on k-nearest neighbors (KNN) regression is proposed. It is suitable for the differential polarization navigation system. This method mainly uses the base station measurements under noisy conditions to train a KNN regression. It contains the relationship between the residual of AoP and the sun position that is difficult to establish an analytical model. Based on the search results of the KNN regression, then we perform a three-step weighted estimation of the residual of AoP for the shipborne mobile station. The simulation shows the effectiveness of this method. After AoP residual estimation and correction, the RMSE of sun azimuth and sun zenith angle estimated from polarization information is less than 0.2°, thus achieving robust polarization navigation under large-scale harsh atmospheric interference within the MABL.