Objectives The positioning accuracy of gravity-aided navigation is closely related to the distribution of the gravitational field in the navigation area. Therefore, the selection of the navigation adaptation area is critical to ensuring the performance of gravity-aided navigation. To ensure high positioning accuracy in gravity-aided navigation, we propose a gravity-aided navigation adaptation area selection algorithm based on a deep gravity field perception model, aiming to improve the accuracy of selecting the navigation adaptation area.

Methods First, the gravity field feature set is constructed using the gray-level co-occurrence matrix, local binary pattern, and the normalized gravity scale proposed in this paper. Second, a deep gravity field perception model is designed, and an inter-layer propagation mechanism is proposed for training the model. Finally, the information from the gravity field feature set is used as input to extract the feature values of the navigation area, and the selected feature values are used to choose the gravity-aided navigation adaptation area. We select a high-precision satellite altimetry-based global ocean gravity field model, and conduct gravity-aided navigation experiments under different conditions for 577 sub-regions (5°×5°each) in the Atlantic and Pacific Oceans. For each sub-region, 100 navigation routes are selected.

Results The experimental results show that the Spearman correlation coefficient between the deep gravity field perception model and the average positioning error ranges from 0.61 to 0.91, while the Kendall correlation coefficient ranges from 0.41 to 0.74. In contrast, the Spearman correlation coefficient between the traditional multi-attribute decision-making method and the average positioning error ranges from 0.22 to 0.70, and the Kendall correlation coefficient ranges from 0.20 to 0.63.

Conclusions The statistical relationship between the feature extraction values of the deep gravity field perception model and the average positioning error, based on a large number of gravity-aided navigation test samples, shows that the features extracted by the proposed model are highly correlated with gravity-aided positioning errors under different conditions. Therefore, the regions selected according to the feature extraction value levels of the deep gravity field perception model correspond to the respective levels of gravity-aided navigation positioning accuracy. This approach improves the accuracy of selecting gravity-aided navigation adaptation areas.