With the development of intelligent waterborne transportation, mining collision avoidance patterns based on spatiotemporal and motion data of ships are crucial for the autonomous navigation of intelligent ships, which requires accurate collision avoidance information under various encounter scenarios. Addressing the existing issues of low precision and false detection in data mining algorithms, this paper proposes a collision avoidance behavior mining model considering encounter scenarios. The model is based on the Automatic Identification System (AIS) and the International Regulations for Preventing Collisions at Sea (COLREGs); it firstly identifies ship collision avoidance turning points by analyzing trajectory curvature with turning and recovering factors. Then, by combining AIS data and the specific navigational environment, it matches the ship encounter pairs and determines the encounter scenarios. Comparative experiments show that the model demonstrates superior accuracy in various scenarios compared to traditional algorithm. Finally, the model was applied to AIS data east of the Yangtze River Estuary, recognizing a total of 827 instances of ship collision avoidance behavior under different encounter scenarios. The case study shows that the model can precisely mine collision avoidance information, laying a solid foundation for future research on autonomous collision avoidance decision making for intelligent ships.

Ship collision avoidance decision-making is a core determinant of navigational safety, and its effectiveness directly governs a ship’s ability to operate safely in a complex environment. Although successive editions of the COLREGs have provided a relatively systematic qualitative description of the principal encounter state and corresponding handling principles, they still lack a quantitative specification of collision avoidance actions. As a result, ship officers must rely heavily on experiential judgment for state recognition and decision-making in real operations, which in turn increases the likelihood of human error-induced failures in avoidance. To address this problem, this study constructs a COLREG-compliant collision avoidance decision azimuth map derived from the collective wisdom of ship officers. Specifically, a two-stage mining process tailored to large-scale AIS data is designed: in the first stage, ship–ship encounter cases are extracted from the full dataset, and, in the second stage, collision avoidance actions are mined from these encounter cases. Subsequently, a decision tree classification model is employed to partition the latent relationships between the relative motion features of two ships and their collision avoidance actions under different encounter scenarios, thereby constructing a data-driven ship collision avoidance decision azimuth map. Finally, taking the eastern coastal waters of China as a case study, the constructed ship collision avoidance action azimuth map is shown to provide scenario-specific guidance for two-ship encounters, offer an objective basis for the quantitative enrichment of COLREGs, and supply a methodological reference for future autonomous ship collision avoidance systems.

Conventional ship-domain models often prescribe boundary geometries empirically or infer them statistically from observed traffic, which may limit their ability to represent the directional variation in modeled encounter risk. This study develops a risk-contour-based ship-domain model using Riemann-sphere projection. Relative position and motion are mapped from the Euclidean encounter plane into a bounded spherical space, where projected relative distance and signed projected closing speed are combined to construct a composite encounter-risk index. A sigmoid transformation is used for the closing-speed component, and softmax-based weights adaptively combine the distance- and speed-related components. For head-on, starboard-side crossing, port-side crossing, and overtaking scenarios, the first positions at which the composite risk reaches a prescribed threshold are extracted for each relative bearing. These scenario-specific first-threshold-crossing points are fused using a pointwise outer-envelope rule, followed by periodic interpolation and smoothing, to reconstruct a directional ship-domain boundary. The reconstructed boundary is evaluated using AIS-derived encounter trajectories from the Florida Straits. The evaluation examines the temporal relationship among predicted domain entry, observed domain entry, and detectable changes in course or speed under a constant-course and constant-speed extrapolation assumption. Under the adopted encounter-screening and high-risk classification criteria, all identified high-risk samples generated a predicted-entry warning, while the predicted-entry warning rate for routine encounters ranged from 30.4% to 40.5%. Among the analyzed encounter trajectories, 17.4% entered the reconstructed ship domain. These results indicate that the proposed procedure can generate a directional and fore-and-aft asymmetric model-based warning contour. However, the reported warning performance is conditional on the selected projection scale, risk parameters, threshold, motion-extrapolation rule, AIS processing procedure, and high-risk labeling criteria. The reconstructed ship domain should therefore be interpreted as a model-based warning contour rather than as a direct measurement of mariners’ psychological perception or a direct indicator of collision probability.