An enhanced UAV GNSS/SINS positioning method with mountain scenario awareness and stochastic model optimization

Abstract To address the degradation of positioning accuracy in unmanned aerial vehicles (UAV) global navigation system and strap-down inertial navigation system (GNSS/SINS) integrated navigation systems operating in complex mountainous environments, this paper proposes a new GNSS/SINS integrated positioning method based on mountain scenario awareness and stochastic model optimization. By analyzing GNSS signal characteristics under three typical mountain scenarios, open mountainous areas, single-sided canyons, and double-sided canyons, a data-driven long short-term memory mountain scenario classification model is developed. Building upon this, a scenario-adaptive GNSS stochastic model is constructed by exploring the correlation between pseudorange errors and carrier-to-noise ratio (CNR), which is subsequently integrated into a GNSS/SINS fusion algorithm within a factor graph optimization (FGO) framework to dynamically refine the information matrix. Through real mountain experiments, the results show that the classification accuracy of the constructed mountain scenario awareness model reaches 90.67%, which is an average increase of 5.11% compared with the support vector machine, multilayer perceptron, and convolutional neural network methods; in a Differential Global Positioning System integration mode, the proposed scenario-adaptive method improves positioning accuracy by 20.9% in an extended kalman filter-based framework and by 23.1% in an FGO-based framework, compared to traditional CNR stochastic models; in real-time kinematic mode, improvements of 20.4% and 13.7% are observed respectively, along with ambiguity resolution success rates increased by 8.7% and 8.9%, which verifies the effectiveness and robustness of the method in complex mountainous environments and provides a new solution for the enhancement of UAV GNSS/SINS integrated navigation in mountainous regions.