Leaning behavior recognition method and device, electronic equipment, readable storage medium and chip
By integrating neural network recognition models and utilizing CNN and Transformer network structures, the system identifies berthing behavior in the trajectory data of individual vessels, solving the regulatory blind spot problem when only one vessel has AIS activated, and realizing automatic identification of vessel berthing behavior.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YIHAILAN (BEIJING) DATA TECH CO LTD
- Filing Date
- 2025-07-30
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technology cannot effectively identify berthing behavior when only one vessel has its AIS activated, creating a regulatory blind spot.
By acquiring trajectory data of two vessels, a fusion neural network recognition model is used. Combining local dwelling features and heading change patterns in the trajectory data, the neural network is trained to identify the berthing behavior of a single vessel. CNN and Transformer network structures are used for feature extraction and global behavioral dependency analysis.
It enables automatic identification of vessel berthing behavior even when only one vessel has its AIS activated, improving the integrity and coverage of maritime supervision.
Smart Images
Figure CN120998065B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of maritime regulatory technology, and more specifically, to a method, apparatus, electronic device, readable storage medium, and chip for identifying docking behavior. Background Technology
[0002] berthing at sea is a key monitoring target for current maritime regulations. Existing technologies typically rely on trajectory data from the Automatic Identification Systems (AIS) of both vessels, using methods such as trajectory alignment and distance calculation between ships to automatically identify berthing. However, in actual regulatory scenarios, some vessels, in order to evade regulatory obligations, will actively turn off their AIS equipment when berthing or transferring goods, resulting in the loss of their trajectory data. In this case, existing identification methods relying on the trajectories of both vessels become completely ineffective and cannot cover the identification needs of berthing when only one vessel has its AIS activated, resulting in a significant blind spot. Summary of the Invention
[0003] The purpose of this invention is to provide a method, apparatus, electronic device, readable storage medium, and chip for identifying berthing behavior, which can solve the problem of not being able to cover the berthing behavior identification needs when only one ship has AIS enabled.
[0004] In view of this, an embodiment of the first aspect of the present invention provides a method for recognizing leaning behavior.
[0005] A second aspect of the present invention provides a device for reclining behavior recognition.
[0006] An embodiment of the third aspect of the present invention provides an electronic device.
[0007] An embodiment of the fourth aspect of the present invention provides a readable storage medium.
[0008] An embodiment of the fifth aspect of the present invention provides a chip.
[0009] To achieve the above objectives, an embodiment of the first aspect of the present invention provides a method for identifying berthing behavior, comprising: acquiring a first travel trajectory of two vessels and corresponding first trajectory data; determining a berthing behavior trajectory segment corresponding to the occurrence of berthing behavior based on the first travel trajectory and first trajectory data of the two vessels; determining a single-vessel trajectory positive sample dataset based on the berthing behavior trajectory segment; acquiring a second travel trajectory of the two vessels and corresponding second trajectory data; determining a normal navigation trajectory segment corresponding to normal vessel travel based on the second travel trajectory and second trajectory data of the two vessels; determining a single-vessel trajectory negative sample dataset based on the normal navigation trajectory segment; training a fusion neural network recognition model based on the single-vessel trajectory positive sample dataset and the single-vessel trajectory negative sample dataset, wherein the fusion neural network recognition model is configured to take the target trajectory data of a single target vessel as input; acquiring the target trajectory data of the single target vessel to be identified; determining the trajectory segment corresponding to the target trajectory data; inputting the target trajectory data into the fusion neural network recognition model to determine the berthing behavior state corresponding to at least one trajectory segment of the single target vessel.
[0010] The berthing behavior recognition method provided by this invention acquires vessel trajectory data from the Automatic Identification System (AIS). Based on the trajectories of two vessels and their relative distance changes, combined with rules such as dwell time, relative distance thresholds, or relative motion trends, trajectory segments indicating obvious berthing behavior between the two vessels are identified as positive sample sources. From the positive sample sources, the trajectory segments corresponding to the berthing behavior of the target vessel are extracted, retaining only the trajectory data of that vessel itself. At least one trajectory segment corresponding to a single vessel exhibiting obvious berthing behavior is used as a single-vehicle trajectory positive sample dataset. Simultaneously, normal navigation trajectory segments of vessels that have not engaged in berthing behavior are collected as negative sample sources. At least one trajectory segment corresponding to a single vessel exhibiting normal navigation behavior is used as a single-vehicle trajectory negative sample dataset. A neural network model is trained using the positive and negative sample sets to determine the fused neural network recognition model. The model analyzes the trajectory data of a single vessel collected by AIS, extracting features such as local dwelling characteristics, speed and / or heading change patterns, etc., and autonomously learns the data features of multiple trajectory segments from the positive and negative sample sets. The trajectory data of a single target vessel is analyzed based on the data characteristics of the positive and negative sample sets, and the result of the judgment on whether the vessel has engaged in berthing behavior is output through a fully connected classifier.
[0011] Understandably, by dividing a ship's navigation trajectory into multiple trajectory segments and marking these segments, positive samples of single-ship trajectories exhibiting berthing behavior can be identified. By combining positive and negative samples, the neural network model can learn the trajectory and parameter features corresponding to berthing behavior. It only needs to input the target ship's trajectory to perform inference and automatically determine whether the ship has engaged in berthing behavior. Even if another ship has its AIS turned off, berthing behavior can still be identified, thus improving the completeness and coverage of maritime behavior supervision.
[0012] In some technical solutions, optionally, the docking behavior trajectory segment corresponding to the docking behavior is determined based on the first travel trajectory and first trajectory data of the two vessels, including: identifying the first vessel and the second vessel in the two vessels; acquiring historical docking data of the vessels; determining a docking behavior distance threshold based on the historical docking data of the vessels; determining multiple trajectory segments of the two vessels; determining the relative distance between the first vessel and the second vessel; when the relative distance is less than the docking behavior distance threshold and the first trajectory data of the first vessel and the first trajectory data of the second vessel change synchronously, determining that there is a docking behavior between the first vessel and the second vessel; and determining at least one trajectory segment corresponding to the docking behavior as a docking behavior trajectory segment.
[0013] In this scheme, trajectory data of ships is collected through AIS. Based on the trajectories of two ships and their relative distance changes, combined with rules such as dwell time, relative distance threshold, or relative motion trend, trajectory segments showing obvious berthing behavior between the two ships are identified as positive sample sources. From the positive sample sources, the trajectory segments of the target ship corresponding to the berthing behavior are extracted, retaining only the trajectory data of that ship itself. At least one trajectory segment showing obvious berthing behavior for a single ship is designated as the berthing behavior trajectory segment.
[0014] Understandably, by transforming abstract behaviors into a quantifiable process for synchronous analysis of distance calculation behaviors, an accurate data source is provided for the learning and training of the fusion neural network recognition model, thereby improving the accuracy of the fusion neural network recognition model in identifying individual ship berthing behaviors.
[0015] In some technical solutions, optionally, determining the berthing behavior state corresponding to at least one trajectory segment of a single target vessel includes: determining the local dwelling characteristics of the trajectory segment corresponding to the target trajectory data by fusing a neural network recognition model; determining data change parameters based on the target trajectory data; determining the behavior dependency parameters of the target trajectory data corresponding to the berthing behavior trajectory segment based on the local dwelling characteristics and the data change parameters; determining the berthing behavior state of the single target vessel within the trajectory segment based on the behavior dependency parameters, wherein the berthing behavior state includes a first behavior state and a second behavior state, the target trajectory data in the first behavior state corresponds to a positive sample dataset, and the target trajectory data in the second behavior state corresponds to a negative sample dataset.
[0016] In this scheme, the fusion neural network recognition model adopts a trajectory behavior model that combines CNN and Transformer network structures. The fusion neural network recognition model uses positive and negative sample datasets of single-ship trajectories as training sets. By analyzing the time-series vector sequences of multiple trajectory segments, the model determines the characteristics of single-ship trajectory segments that exhibit docking behavior and those that do not.
[0017] The CNN module in the fusion neural network recognition model is used to extract multiple navigation data change parameters from the ship trajectory data, such as local dwelling features, speed changes, and heading changes. The Transformer network structure in the fusion neural network recognition model is used to capture the global behavioral dependencies of trajectory segments in the time dimension. The global behavioral dependencies are used to determine the dependency of a single ship trajectory segment on a positive or negative sample of a single ship trajectory within a time window. A fully connected classifier outputs whether the single ship to be identified has engaged in berthing behavior. If it is determined that the single ship to be identified has not engaged in berthing behavior, the berthing behavior state of the single ship to be identified is determined as the second behavioral state; if it is determined that the single ship to be identified has engaged in berthing behavior, the berthing behavior state of the single ship to be identified is determined as the first behavioral state.
[0018] Understandably, this involves inputting the target trajectory data of a single vessel into a fusion neural network recognition model. By relying on spatial proximity and behavioral synchronization, the behavioral dependency between the target trajectory data of a single vessel and the temporal vector sequences corresponding to positive and negative samples is determined. This transforms the identification of a single vessel's berthing behavior into a quantifiable output, providing full-scenario requirements for vessel berthing behavior recognition. It enables the identification of berthing behavior even when only one vessel has AIS activated, thereby improving the completeness of maritime behavior supervision.
[0019] In some technical solutions, optionally, a single-ship trajectory negative sample dataset is determined based on normal navigation trajectory segments, including: determining multiple normal navigation trajectory segments of the first and second ships; determining the second trajectory data corresponding to the normal navigation trajectory segments; determining a time-series vector sequence based on the second trajectory data corresponding to the normal navigation trajectory segments; and determining a single-ship trajectory negative sample dataset based on multiple time-series vector sequences.
[0020] In this scheme, normal navigation trajectory segments of vessels that did not engage in berthing are collected as negative samples, forming a single-vessel trajectory negative sample dataset. Each sample is a time-series vector sequence.
[0021] Ship trajectory data is collected using AIS. Based on the trajectories of two ships and their relative distance changes, combined with rules such as dwell time, relative distance thresholds, or relative motion trends, ships that did not engage in berthing are identified. The navigation trajectory segments of these ships are used as negative sample sources. From the negative sample sources, a second navigation trajectory for a single ship is determined, and multiple trajectory segments within this second navigation trajectory are identified. These multiple trajectory segments are considered normal navigation trajectory segments. The negative sample dataset is then determined based on the trajectory data corresponding to these multiple normal navigation trajectory segments.
[0022] In some technical solutions, optionally, determining the trajectory segment corresponding to the target trajectory data includes: cleaning the target trajectory data to remove abnormal data points, determining the cleaned target trajectory data, the target trajectory data including multiple data points; determining the target vessel's target travel trajectory based on the cleaned target trajectory data; segmenting the target travel trajectory to determine multiple trajectory segments including complete speed change cycles, the data points within the trajectory segments having the same time interval.
[0023] In this solution, after obtaining the target trajectory data of the target vessel from AIS, the target trajectory data is preprocessed, including cleaning, segmenting, outlier removal, and resampling. This preprocessing provides standard input data and ensures the effectiveness of the feature data extraction by the fusion neural network recognition model.
[0024] In some technical solutions, the trajectory data may optionally include first trajectory data, second trajectory data, and target trajectory data; the data fields corresponding to the trajectory data include: timestamp, longitude, latitude, speed, heading, time interval, and behavior rhythm.
[0025] In this scheme, the first trajectory data, the second trajectory data, and the target trajectory data are all trajectory data of the corresponding ships collected in the AIS system. The data fields corresponding to the trajectory data include, but are not limited to: timestamp, longitude, latitude, speed (SOG), heading (COG), time interval or behavior rhythm (delta_t).
[0026] A second aspect of the present invention provides a docking behavior recognition device, comprising: a positive sample module for acquiring a first travel trajectory of two vessels and corresponding first trajectory data; determining a docking behavior trajectory segment corresponding to the occurrence of docking behavior based on the first travel trajectory and first trajectory data of the two vessels; determining a single vessel trajectory positive sample dataset based on the docking behavior trajectory segment; a negative sample module for acquiring a second travel trajectory of two vessels and corresponding second trajectory data; determining a normal navigation trajectory segment corresponding to normal vessel travel based on the second travel trajectory and second trajectory data of the two vessels; determining a single vessel trajectory negative sample dataset based on the normal navigation trajectory segment; a model training module for training a fusion neural network recognition model based on the single vessel trajectory positive sample dataset and the single vessel trajectory negative sample dataset, wherein the fusion neural network recognition model is configured to take the target trajectory data of a single target vessel as input; a data acquisition module for acquiring the target trajectory data of the single target vessel to be identified; a segment determination module for determining the trajectory segment corresponding to the target trajectory data; and a docking recognition module for inputting the target trajectory data into the fusion neural network recognition model to determine the docking behavior state corresponding to at least one trajectory segment of the single target vessel.
[0027] An embodiment of the third aspect of this application provides an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the docking behavior recognition method as described in the first aspect.
[0028] An embodiment of the fourth aspect of this application provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the docking behavior recognition method as described in the first aspect.
[0029] An embodiment of the fifth aspect of this application provides a chip including a processor and a communication interface, the communication interface and the processor being coupled together, the processor being used to run a program or instructions to implement the steps of the docking behavior recognition method as described in the first aspect.
[0030] Additional aspects and advantages of the technical solutions of the present invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description
[0031] Figure 1 A flowchart illustrating a method for recognizing leaning behavior according to an embodiment of this application is shown; Figure 2 A partial flowchart of a method for recognizing leaning behavior according to an embodiment of this application is shown; Figure 3 A partial flowchart of a method for recognizing leaning behavior according to an embodiment of this application is shown; Figure 4 A partial flowchart of a method for recognizing leaning behavior according to an embodiment of this application is shown; Figure 5 A partial flowchart of a method for recognizing leaning behavior according to an embodiment of this application is shown; Figure 6 A schematic block diagram of a leaning behavior recognition device according to an embodiment of this application is shown; Figure 7 A schematic block diagram of an electronic device according to an embodiment of this application is shown; Figure 8 A flowchart illustrating a method for recognizing leaning behavior according to an embodiment of this application is shown.
[0032] in, Figure 6 and Figure 7 The correspondence between the reference numerals and component names in the attached drawings is as follows: 900: Dock behavior recognition device; 902: Positive sample module; 904: Negative sample module; 906: Model training module; 908: Data acquisition module; 910: Segment determination module; 912: Dock recognition module; 1000: Electronic device; 1109: Memory; 1110: Processor. Detailed Implementation
[0033] To better understand the above-described objectives, features, and advantages of the embodiments of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.
[0034] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, embodiments of the invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below.
[0035] The following is in conjunction with the appendix Figures 1 to 8 The present application provides a detailed description of the method, apparatus, electronic device, readable storage medium, and chip for identifying leaning behavior through specific embodiments and application scenarios.
[0036] like Figure 1 As shown, this embodiment provides a method for recognizing leaning behavior, including: Step S100: Obtain the first travel trajectory of the two ships and the corresponding first trajectory data; Step S102: Determine the docking behavior trajectory segment corresponding to the docking behavior based on the first driving trajectory and first trajectory data of the two vessels; Step S104: Determine the positive sample dataset of single-ship trajectory based on the docking behavior trajectory segments; Step S106: Obtain the second travel trajectory of the two ships and the corresponding second trajectory data; Step S108: Determine the normal navigation trajectory segment corresponding to the normal navigation of the vessel based on the second driving trajectory and second trajectory data of the two vessels; Step S110: Determine the negative sample dataset of a single ship's trajectory based on the normal navigation trajectory segment; Step S112: Train a fusion neural network recognition model based on the single ship trajectory positive sample dataset and the single ship trajectory negative sample dataset. The fusion neural network recognition model is configured to take the target trajectory data of a single target ship as input. Step S114: Obtain the target trajectory data of the single target vessel to be identified; Step S116: Determine the trajectory segment corresponding to the target trajectory data; Step S118: Input the target trajectory data into the fusion neural network recognition model to determine the berthing behavior state corresponding to at least one trajectory segment of a single target vessel.
[0037] The berthing behavior recognition method provided by this invention acquires vessel trajectory data from the Automatic Identification System (AIS). Based on the trajectories of two vessels and their relative distance changes, combined with rules such as dwell time, relative distance thresholds, or relative motion trends, trajectory segments indicating obvious berthing behavior between the two vessels are identified as positive sample sources. From the positive sample sources, the trajectory segments corresponding to the berthing behavior of the target vessel are extracted, retaining only the trajectory data of that vessel itself. At least one trajectory segment corresponding to a single vessel exhibiting obvious berthing behavior is used as a single-vehicle trajectory positive sample dataset. Simultaneously, normal navigation trajectory segments of vessels that have not engaged in berthing behavior are collected as negative sample sources. At least one trajectory segment corresponding to a single vessel exhibiting normal navigation behavior is used as a single-vehicle trajectory negative sample dataset. A neural network model is trained using the positive and negative sample sets to determine the fused neural network recognition model. The model analyzes the trajectory data of a single vessel collected by AIS, extracting features such as local dwelling characteristics, speed and / or heading change patterns, etc., and autonomously learns the data features of multiple trajectory segments from the positive and negative sample sets. The trajectory data of a single target vessel is analyzed based on the data characteristics of the positive and negative sample sets, and the result of the judgment on whether the vessel has engaged in berthing behavior is output through a fully connected classifier.
[0038] Understandably, by dividing a ship's navigation trajectory into multiple trajectory segments and marking these segments, positive samples of single-ship trajectories exhibiting berthing behavior can be identified. By combining positive and negative samples, the neural network model can learn the trajectory and parameter features corresponding to berthing behavior. It only needs to input the target ship's trajectory to perform inference and automatically determine whether the ship has engaged in berthing behavior. Even if another ship has its AIS turned off, berthing behavior can still be identified, thus improving the completeness and coverage of maritime behavior supervision.
[0039] Specifically, AIS (Automatic Information System) is used to collect the trajectories of two ships as samples for docking behavior in the neural network model. Among the collected trajectories of multiple ships, the two ships that docked are identified and used as the docking behavior samples. A time window of a certain length is defined, which can be preset within the system to reduce data processing. This window is used to determine the two segments of the ship's trajectory from the starting point to the ending point of the voyage. After determining the time window and the trajectories of the two ships that docked, based on rules such as dwell time, relative distance threshold, and relative motion trend, the trajectory segments within the time window that exhibit obvious docking behavior are identified. These trajectory segments with obvious docking behavior are labeled and used as positive sample sources.
[0040] Furthermore, after identifying trajectory segments exhibiting clear berthing behavior, the two vessel trajectories corresponding to the two vessels are merged into a single berthing behavior trajectory segment, i.e., treated as an isolated trajectory segment for a single vessel. Within this berthing behavior trajectory segment, berthing occurs between the two vessels, their relative motion trends are identical, and the relative distance between them is less than a berthing behavior distance threshold. This berthing behavior trajectory segment is then identified as a positive sample of the single-vessel trajectory.
[0041] Furthermore, in the navigation trajectories of multiple vessels collected by AIS, the navigation trajectories of multiple vessels that did not engage in berthing are identified. These trajectories corresponding to the absence of berthing are defined as normal navigation trajectories. Then, using a time window approach, multiple normal navigation trajectory segments are identified within these normal navigation trajectories. These normal navigation trajectory segments are labeled and used as negative samples of individual vessel trajectories.
[0042] Furthermore, the fusion neural network recognition model includes a cascaded structure of convolutional neural networks (CNN) and deep learning models (Transformer) network structures.
[0043] The fusion neural network recognition model employs a trajectory behavior model that combines CNN and Transformer network structures. Using both positive and negative sample datasets of single-ship trajectories as training sets, the model analyzes the time-series vector sequences of multiple trajectory segments to determine the characteristics of single-ship trajectory segments exhibiting or lacking aggregating behavior.
[0044] The CNN module in the fusion neural network recognition model is used to extract multiple navigation data change parameters such as local dwelling features, speed changes, and heading changes from the ship trajectory data. The Transformer network structure in the fusion neural network recognition model is used to capture the global behavioral dependency of the trajectory segment in the time dimension. The global behavioral dependency is used to determine the dependency of a single ship trajectory segment on a positive or negative sample of the single ship trajectory within a time window. The global representation vector ([CLS] token) is used as the overall semantic representation of the single ship trajectory segment, and the fully connected classifier outputs whether the single ship to be identified has berthing behavior.
[0045] For example, the berthing behavior status output by the fully connected classifier corresponds to a binary classification label. When the output result is 0, it is determined that the individual vessel to be identified does not engage in berthing behavior; when the output result is 1, it is determined that the individual vessel to be identified engages in berthing behavior.
[0046] For example, the Transformer network structure includes an encoder and a decoder. The encoder consists of multiple stacked encoder layers, each containing the following sub-layers: a multi-head attention mechanism: by calculating dot product attention, it captures the dependency between any two elements in the input sequence, i.e., the dependency between the single ship trajectory to be identified and the positive and negative samples of the single ship trajectory; a feedforward neural network: it performs linear transformation and activation function processing on the output of the self-attention layer to enhance the model's expressive power; residual connections and layer normalization: help stabilize model training and avoid the gradient vanishing problem. The decoder structure is similar to the encoder, but adds a masked self-attention mechanism to prevent information leakage. Through the multi-head attention mechanism, the trajectory segment of a single ship in the input is divided into multiple subspaces for parallel processing. The trajectory segment of a single ship is multi-dimensional time-series data, improving the model's ability to capture complex relationships.
[0047] Furthermore, both the positive and negative sample datasets of single-ship trajectories consist of isolated trajectory segments of a single ship, without relying on trajectory data from other ships.
[0048] In some embodiments, optionally, such as Figure 2 As shown, step S102: Based on the first travel trajectory and first trajectory data of the two vessels, determine the docking behavior trajectory segment corresponding to the docking behavior, including: Step S1020: Determine the first and second vessels in the two vessels; Step S1022: Obtain historical data on vessel berthing; Step S1024: Determine the distance threshold for berthing behavior based on historical ship berthing data; Step S1026: Determine multiple trajectory segments of the two ships; Step S1028: Determine the relative distance between the first vessel and the second vessel; Step S1030: When the relative distance is less than the berthing distance threshold and the first trajectory data of the first vessel and the first trajectory data of the second vessel change synchronously, it is determined that there is a berthing behavior between the first vessel and the second vessel. Step S1032: Determine at least one trajectory segment corresponding to the docking behavior as the docking behavior trajectory segment.
[0049] In this embodiment, trajectory data of ships is collected via AIS. Based on the trajectories of two ships and their relative distance changes, combined with rules such as dwell time, relative distance threshold, or relative motion trend, trajectory segments showing obvious berthing behavior between the two ships are identified as positive sample sources. From the positive sample sources, the trajectory segments of the target ship corresponding to the berthing behavior are extracted, retaining only the trajectory data of that ship itself. At least one trajectory segment showing obvious berthing behavior for a single ship is designated as the berthing behavior trajectory segment.
[0050] Understandably, by transforming abstract behaviors into a quantifiable process for synchronous analysis of distance calculation behaviors, an accurate data source is provided for the learning and training of the fusion neural network recognition model, thereby improving the accuracy of the fusion neural network recognition model in identifying individual ship berthing behaviors.
[0051] Specifically, from the trajectory data of multiple vessels collected by AIS, the two vessels involved in the berthing behavior are identified as the first vessel and the second vessel, and both vessels are marked. From the first travel trajectory and the first trajectory data of the first and second vessels, the complete travel trajectories of the first and second vessels are determined. The continuous complete travel trajectory is then divided into multiple trajectory segments of fixed duration or fixed length according to time windows.
[0052] Furthermore, multiple known vessel berthing events are identified from the AIS database, and the vessel spacing distance at the time of berthing is determined based on these events. The distribution of these vessel spacing distances is then determined, and the average of these distances is used as the berthing distance threshold.
[0053] The trajectory data within multiple trajectory segments are analyzed separately. The trajectory data includes timestamps, longitude, latitude, speed, heading, time intervals, and behavioral rhythm. The relative distance between the first and second vessels is determined based on their longitude and latitude data. This relative distance is then compared to a berthing distance threshold. When the relative distance between the first and second vessels is less than or equal to the berthing distance threshold, the distance between them is considered to meet the berthing distance requirement. If the distance between the first and second vessels meets the berthing distance requirement, their speed and heading are analyzed separately. When the first and second vessels simultaneously change speed and / or simultaneously change heading, berthing is determined to have occurred within the trajectory segment, and the corresponding trajectory segment is identified as the berthing behavior trajectory segment.
[0054] Furthermore, the hitchhiking behavior trajectory segments are labeled, and their corresponding binary labels are determined. For example, the binary label is 1, and the labeled hitchhiking behavior trajectory segments are used as positive sample data of single ship trajectory.
[0055] Furthermore, if the relative distance between the first vessel and the second vessel remains less than the berthing distance threshold, it is determined that the first vessel and the second vessel meet the distance requirement corresponding to the berthing action. The duration is determined based on the duration corresponding to the berthing action in multiple vessel berthing events, and the duration is the average of the durations corresponding to the berthing action in multiple vessel berthing events.
[0056] Furthermore, if the distance between the first vessel and the second vessel meets the distance requirement corresponding to the berthing behavior, and the first vessel and the second vessel continue to maintain a low speed or a stationary state, it is determined that the first vessel and the second vessel have engaged in berthing behavior within the trajectory segment, and the corresponding trajectory segment is determined as the berthing behavior trajectory segment.
[0057] Furthermore, the distance threshold for the aggregating behavior is dynamically updated, and the dynamic update time interval can be set in the AIS system.
[0058] Furthermore, by determining a time window, the trajectory data of the first and second vessels are analyzed within the time window to identify the trajectory segments within the time window that show obvious berthing behavior.
[0059] The window length and step size of the time window are preset values, determined by performing sliding window interception on the real-time AIS data stream.
[0060] Furthermore, the ship trajectory data within the time window is resampled to a fixed frequency, for example, 10Hz. After resampling to the fixed frequency, positive samples of the individual ship trajectory are determined based on at least one trajectory segment.
[0061] Furthermore, each sample in the single-ship trajectory positive sample dataset is represented as a time-series vector sequence, including longitude, latitude, speed, heading, and behavioral rhythm.
[0062] In some embodiments, optionally, such as Figure 3 As shown, step S118: Determine the berthing behavior state corresponding to at least one trajectory segment of a single target vessel, including: Step S1180: Determine the local dwell characteristics of the trajectory segment corresponding to the target trajectory data by using a fusion neural network recognition model; Step S1182: Determine the data change parameters based on the target trajectory data; Step S1184: Determine the behavior dependency parameters of the target trajectory data corresponding to the aggregating behavior trajectory segment based on local dwell characteristics and data change parameters; Step S1186: Determine the berthing behavior state of a single target vessel within the trajectory segment based on the behavior dependency parameters. The berthing behavior state includes a first behavior state and a second behavior state. The target trajectory data under the first behavior state corresponds to the positive sample dataset, and the target trajectory data under the second behavior state corresponds to the negative sample dataset.
[0063] In this embodiment, the fusion neural network recognition model adopts a trajectory behavior model that combines CNN and Transformer network structures. The fusion neural network recognition model uses the positive sample dataset and the negative sample dataset of single-ship trajectory as training sets. The model analyzes the time-series vector sequences of multiple trajectory segments to determine the features of single-ship trajectory segments that exhibit docking behavior and the features of single-ship trajectory segments that do not exhibit docking behavior.
[0064] The CNN module in the fusion neural network recognition model is used to extract multiple navigation data change parameters from the ship trajectory data, such as local dwelling features, speed changes, and heading changes. The Transformer network structure in the fusion neural network recognition model is used to capture the global behavioral dependencies of trajectory segments in the time dimension. The global behavioral dependencies are used to determine the dependency of a single ship trajectory segment on a positive or negative sample of a single ship trajectory within a time window. A fully connected classifier outputs whether the single ship to be identified has engaged in berthing behavior. If it is determined that the single ship to be identified has not engaged in berthing behavior, the berthing behavior state of the single ship to be identified is determined as the second behavioral state; if it is determined that the single ship to be identified has engaged in berthing behavior, the berthing behavior state of the single ship to be identified is determined as the first behavioral state.
[0065] Understandably, this involves inputting the target trajectory data of a single vessel into a fusion neural network recognition model. By relying on spatial proximity and behavioral synchronization, the behavioral dependency between the target trajectory data of a single vessel and the temporal vector sequences corresponding to positive and negative samples is determined. This transforms the identification of a single vessel's berthing behavior into a quantifiable output, providing full-scenario requirements for vessel berthing behavior recognition. It enables the identification of berthing behavior even when only one vessel has AIS activated, thereby improving the completeness of maritime behavior supervision.
[0066] Furthermore, the local dwell characteristics are determined by the CNN module in the neural network recognition model, which detects at least three types of key features in the target trajectory segment. The key features include speed dwell characteristics, heading disturbance characteristics, and positional stagnation characteristics.
[0067] Furthermore, speed and heading changes are determined using the flight speed and heading data, respectively. Speed changes are used to detect abnormal deceleration behavior, while heading changes are used to determine turning behavior that deviates from the normal flight path.
[0068] Furthermore, the Transformer module integrates local dwelling features, parameters corresponding to speed changes, and parameters corresponding to heading changes to construct a cross-modal dependency matrix. A multi-head attention mechanism is then used to determine the global behavioral dependency of the trajectory data of the individual vessel to be identified on either a positive or negative sample dataset of the single-vehicle trajectory. Based on this global behavioral dependency, the berthing behavior state of the individual vessel to be identified is determined.
[0069] In some embodiments, optionally, such as Figure 4 As shown, step S110: Determine the negative sample dataset of a single ship's trajectory based on the normal navigation trajectory segment, including: Step S1100: Determine multiple normal navigation trajectory segments of the first and second vessels; Step S1102: Determine the second trajectory data corresponding to the normal navigation trajectory segment; Step S1104: Determine the time-series vector sequence based on the second trajectory data corresponding to the normal navigation trajectory segment; Step S1106: Determine the negative sample dataset of a single ship trajectory based on multiple time-series vector sequences.
[0070] In this embodiment, normal navigation trajectory segments of vessels that did not engage in docking are collected as negative samples to form a single-vessel trajectory negative sample dataset. Each sample is a time-series vector sequence.
[0071] Ship trajectory data is collected using AIS. Based on the trajectories of two ships and their relative distance changes, combined with rules such as dwell time, relative distance thresholds, or relative motion trends, ships that did not engage in berthing are identified. The navigation trajectory segments of these ships are used as negative sample sources. From the negative sample sources, a second navigation trajectory for a single ship is determined, and multiple trajectory segments within this second navigation trajectory are identified. These multiple trajectory segments are considered normal navigation trajectory segments. The negative sample dataset is then determined based on the trajectory data corresponding to these multiple normal navigation trajectory segments.
[0072] Furthermore, in the normal navigation trajectory segment, the relative distance between the target vessel and at least one surrounding vessel is greater than the berthing distance threshold, and the target vessel's navigation speed does not change abruptly in multiple trajectory segments.
[0073] For example, the time-series vector sequence is represented as [lon, lat, sog, cog, delta_t]. Where lon is longitude, lat is latitude, sog is speed, cog is heading, and delta_t is behavior rhythm.
[0074] Furthermore, in the negative sample dataset, the specific location of the ship is determined based on the longitude and latitude in the trajectory data. The distance between the ship and multiple ports is then determined based on the ship's specific location to avoid berthing interference. Trajectory segments in the negative sample dataset where the distance between the ship and the port is less than a preset value are deleted.
[0075] Furthermore, the normal navigation trajectory segments are labeled, and their corresponding binary labels are determined. For example, if the binary label is 0, the labeled normal navigation trajectory segments are used as negative sample data of a single ship trajectory.
[0076] Furthermore, each individual ship trajectory negative sample has an independent code used to determine the ship's independent behavior.
[0077] In some embodiments, optionally, such as Figure 5 As shown, step S116: Determine the trajectory segment corresponding to the target trajectory data, including: Step S1160: Clean the target trajectory data to remove abnormal data points, and determine the cleaned target trajectory data, which includes multiple data points; Step S1162: Determine the target vessel's trajectory based on the cleaned target trajectory data; Step S1164: Segment the target driving trajectory to determine multiple trajectory segments that include complete speed change cycles, with data points within the trajectory segments having the same time interval.
[0078] In this embodiment, after obtaining the target trajectory data of the target vessel from AIS, the target trajectory data is preprocessed to complete preprocessing operations such as cleaning the target trajectory data, segmenting the trajectory, removing outliers, and resampling the trajectory, providing standard input data and ensuring the effectiveness of the feature data extracted by the fusion neural network recognition model.
[0079] Specifically, outlier data points include position drift points, velocity abrupt change points, and timestamp errors. Position drift points are defined as those where the displacement velocity is greater than the ship's maximum speed or less than its berthing speed; velocity abrupt change points are those where the speed change is greater than the average value and occurs outside of port operations; and timestamp errors correspond to timestamps that are either out of order or abnormal. Multiple outlier data points in the target trajectory data are cleaned to determine the cleaned target trajectory data. By removing outlier data, the velocities of multiple trajectory segments corresponding to the target trajectory data change continuously. The sampling frequency of data points in the trajectory segments is uniformly set to a fixed time interval to meet the model input tensor requirements.
[0080] Furthermore, the speed change characteristics in the trajectory segment include continuous acceleration to reach the peak speed point and continuous deceleration to reach the minimum speed point, so as to form a closed-loop speed change pattern in a trajectory segment.
[0081] Furthermore, the removed outlier data points are supplemented by linear interpolation or median filtering smoothing to make the data points in the trajectory data continuous.
[0082] Furthermore, before determining the positive sample dataset of a single ship trajectory, the trajectory data of the first and second ships are preprocessed by cleaning, segmenting, removing outliers, and resampling to form continuous and stable trajectory segments.
[0083] Furthermore, before determining the negative sample dataset of a single ship trajectory, the trajectory data of the target ship is preprocessed by cleaning, segmenting, removing outliers, and resampling to form continuous and stable trajectory segments.
[0084] In some embodiments, the trajectory data may optionally include first trajectory data, second trajectory data, and target trajectory data; the data fields corresponding to the trajectory data include: timestamp, longitude, latitude, speed, heading, time interval, and behavior rhythm.
[0085] In this embodiment, the first trajectory data, the second trajectory data, and the target trajectory data are all trajectory data of the corresponding ships collected in the AIS system. The data fields corresponding to the trajectory data include, but are not limited to: timestamp, longitude, latitude, speed (SOG), heading (COG), time interval or behavior rhythm (delta_t).
[0086] Furthermore, timestamps are the basis for determining ship trajectory segments and aligning ship trajectories. By using timestamps, the collection time corresponding to multiple data points in a trajectory segment is determined, so as to form a time series.
[0087] Furthermore, longitude and latitude are floating-point numbers. The geospatial coordinates of the ship are determined by longitude and latitude to generate a ship position change vector. Thus, the sailing trajectories of multiple ships are determined based on the ship position change vector and timestamps.
[0088] Furthermore, the speed is the ship's speed relative to the ground. By determining the ship's low-speed loitering characteristics and speed change rate, the speed changes of the ship during berthing can be monitored.
[0089] Furthermore, the heading is the ship's heading angle relative to the ground. By identifying the ship's heading disturbances through the heading, abnormal turning when the ship engages in berthing behavior can be monitored.
[0090] Furthermore, the time interval is the time difference between adjacent data points in the trajectory segment.
[0091] Furthermore, the behavioral rhythm is determined based on speed and heading, and is used to determine the data changes of the vessel over a period of time. When the vessel's speed and heading are both normal, the vessel's behavioral rhythm should be normal; when the vessel's speed and heading deviate or become abnormal, the vessel's behavioral rhythm is abnormal, and the vessel's trajectory data is automatically input into the fusion neural network recognition model for berthing behavior identification, thereby achieving automatic monitoring and alarm of vessel berthing behavior.
[0092] In one specific embodiment, optionally, such as Figure 8 As shown, the method for recognizing leaning behavior includes: Step S200: AIS trajectory data acquisition and preprocessing; Step S202: Identification of docking behavior samples based on dual-ship trajectories; Step S204: Construction of single-ship trajectory samples; Step S206: Neural network model design; Step S208: Model training and validation; Step S210: Deploy real-time docking behavior recognition.
[0093] The AIS trajectory data acquisition and preprocessing includes: acquiring spatiotemporal trajectory data of the target vessel from the Automatic Identification System (AIS), with data fields including but not limited to: timestamp, longitude, latitude, speed (SOG), heading (COG), time interval or behavior rhythm (delta_t), etc. Preprocessing operations such as cleaning, trajectory segmentation, outlier removal, and trajectory resampling are then performed on the trajectory data to construct continuous and stable trajectory segments.
[0094] The identification of docking behavior samples based on dual-ship trajectories includes: utilizing the trajectories of two ships and their relative distance changes within a certain time window, combined with rules such as docking time, relative distance threshold, and relative motion trend, to identify trajectory segments exhibiting obvious docking behavior as positive sample sources. Example judgment conditions include: both ships maintaining a low speed or stationary state for a certain period; the distance between them consistently being less than a set threshold (e.g., 50 meters); and behavioral patterns such as synchronized speed changes or synchronized course veergence.
[0095] The construction of single-ship trajectory samples includes: extracting the trajectory segments of the target vessels corresponding to the aforementioned berthing behavior, retaining only the trajectory information of the vessel itself, and constructing single-ship trajectory training samples. Simultaneously, normal navigation trajectory segments of vessels that did not berth are collected as negative samples, forming a positive and negative sample set. Each sample is represented as a time-series vector sequence, including features such as [lon, lat, sog, cog, delta_t].
[0096] The neural network model design includes a trajectory behavior recognition model that integrates a Convolutional Neural Network (CNN) and a deep learning model (Transformer). The CNN module is used to extract local dwell features, speed / heading change patterns, etc., in the trajectory; the Transformer module is used to capture the global behavioral dependencies of the trajectory segment in the time dimension; a global representation vector ([CLS] token) is used as the overall semantic representation of the trajectory segment, and a fully connected classifier is used to output whether there is a docking behavior.
[0097] For example: Model input: shape = (batch_size, sequence_length, 5); Model output: binary labels (0 indicates no overlap, 1 indicates overlap).
[0098] Model training and validation include: supervised training of the model using the aforementioned positive and negative samples. During training, the following are introduced: CrossEntropy Loss; data augmentation (trajectory perturbation, noise simulation); evaluation metrics include: accuracy, recall, and balanced F-score. Cross-validation and sliding window testing are used to ensure the model's generalization ability and robustness.
[0099] The deployment of real-time berthing behavior recognition includes: deploying a trained neural network model to the maritime supervision platform, which can extract the target vessel trajectory segment from the real-time AIS data stream and perform inference to automatically determine whether berthing behavior exists, even if the other vessel turns off its AIS.
[0100] like Figure 6As shown in the figure, this application embodiment also provides a docking behavior recognition device 900, including: a positive sample module 902, used to acquire the first driving trajectory of two ships and the corresponding first trajectory data; determine the docking behavior trajectory segment corresponding to the docking behavior based on the first driving trajectory of the two ships and the first trajectory data; determine the single ship trajectory positive sample dataset based on the docking behavior trajectory segment; a negative sample module 904, used to acquire the second driving trajectory of the two ships and the corresponding second trajectory data; determine the normal navigation trajectory segment corresponding to the normal driving of the ship based on the second driving trajectory of the two ships and the second trajectory data; determine the single ship trajectory negative sample dataset based on the normal navigation trajectory segment; a model training module 906, used to train a fusion neural network recognition model based on the single ship trajectory positive sample dataset and the single ship trajectory negative sample dataset, wherein the fusion neural network recognition model is configured to take the target trajectory data of a single target ship as input; a data acquisition module 908, used to acquire the target trajectory data of the single target ship to be identified; a segment determination module 910, used to determine the trajectory segment corresponding to the target trajectory data; and a docking recognition module 912, used to input the target trajectory data into the fusion neural network recognition model to determine the docking behavior state corresponding to at least one trajectory segment of the single target ship.
[0101] By dividing the ship's navigation trajectory into multiple trajectory segments using the berthing behavior recognition device 900, and marking these segments, positive samples of single-ship trajectories containing berthing behavior are identified. By combining positive and negative samples, the neural network model can learn the trajectory features and parameter features corresponding to berthing behavior. It only needs to input the target ship's trajectory to perform inference and automatically determine whether the ship has engaged in berthing behavior. Even if another ship has its AIS turned off, berthing behavior can still be identified, improving the completeness and coverage of maritime behavior supervision.
[0102] like Figure 7 As shown, this application embodiment also provides an electronic device 1000, including a processor 1110, a memory 1109, and a program or instructions stored in the memory 1109 and executable on the processor 1110. When the program or instructions are executed by the processor 1110, they implement the various processes of the above-described embodiments of the docking behavior recognition method and achieve the same technical effect. To avoid repetition, they will not be described again here.
[0103] Optionally, the processor 1110 is used to acquire the first travel trajectory of the two ships and the corresponding first trajectory data; Optionally, the processor 1110 is also used to determine the docking behavior trajectory segment corresponding to the docking behavior based on the first driving trajectory and first trajectory data of the two ships. Optionally, the processor 1110 is also used to determine a positive sample dataset of a single vessel trajectory based on the berthing behavior trajectory segment; Optionally, the processor 1110 is also used to determine a positive sample dataset of a single vessel trajectory based on the berthing behavior trajectory segment; Optionally, the processor 1110 is also used to determine the normal navigation trajectory segment corresponding to the normal navigation of the ship based on the second driving trajectory and second trajectory data of the two ships. Optionally, the processor 1110 is also used to determine a negative sample dataset of a single ship trajectory based on a normal navigation trajectory segment; Optionally, the processor 1110 is also configured to train a fusion neural network recognition model based on a single ship trajectory positive sample dataset and a single ship trajectory negative sample dataset, wherein the fusion neural network recognition model is configured to take the target trajectory data of a single target ship as input. Optionally, the processor 1110 is also used to acquire target trajectory data of a single target vessel to be identified; Optionally, the processor 1110 is also configured to determine the trajectory segment corresponding to the target trajectory data; Optionally, the processor 1110 is also configured to input the target trajectory data into the fusion neural network recognition model to determine the berthing behavior state corresponding to at least one trajectory segment of a single target vessel.
[0104] The memory 1109 can be used to store software programs and various data. The memory 1109 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 1109 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 1109 in this embodiment includes, but is not limited to, these and any other suitable types of memory.
[0105] This application also provides a readable storage medium storing a program or instructions. When executed by a processor, the program or instructions implement the various processes of the above-described docking behavior recognition method embodiments and achieve the same technical effects. To avoid repetition, these will not be described again here. Furthermore, the readable storage medium improves the data storage capacity and data processing speed of the docking behavior recognition method in this application.
[0106] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing, but is not limited thereto. A non-exhaustive list of more specific examples of computer-readable storage media includes: portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital universal disk (DVD), memory cards, floppy disks, encoding mechanical devices (e.g., punched cards or grooves with raised structures for recording instructions), and any suitable combination of the foregoing. The computer-readable storage medium used herein should not be construed as the transmission of signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media, or electrical signals transmitted through wires.
[0107] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.
[0108] This application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described docking behavior recognition method embodiments, and can achieve the same technical effect. To avoid repetition, it will not be described again here. In addition, the chip improves the data processing speed corresponding to the docking behavior recognition method in this application.
[0109] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.
[0110] In this invention, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance; the term "multiple" refers to two or more unless otherwise explicitly defined. The terms "install," "connect," "link," and "fix" should be interpreted broadly. For example, "connect" can be a fixed connection, a detachable connection, or an integral connection; "link" can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0111] In the description of this invention, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or unit referred to must have a specific orientation or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0112] In the description of this specification, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0113] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for recognizing leaning behavior, characterized in that, include: Obtain the first travel trajectories of the two vessels and the corresponding first trajectory data; Based on the first travel trajectory and the first trajectory data of the two vessels, the docking behavior trajectory segment corresponding to the docking behavior is determined; Determine the positive sample dataset of single-ship trajectory based on the aforementioned docking behavior trajectory segments; Obtain the second travel trajectory of the two vessels and the corresponding second trajectory data; Based on the second travel trajectory and second trajectory data of the two vessels, the normal navigation trajectory segment corresponding to the normal travel of the vessels is determined; Determine the negative sample dataset of a single ship's trajectory based on the normal navigation trajectory segment; A fusion neural network recognition model is trained based on the single ship trajectory positive sample dataset and the single ship trajectory negative sample dataset, wherein the fusion neural network recognition model is configured to take the target trajectory data of a single target ship as input. Acquire the target trajectory data of the single target vessel to be identified; Determine the trajectory segment corresponding to the target trajectory data; The target trajectory data is input into the fusion neural network recognition model to determine the berthing behavior state corresponding to at least one trajectory segment of the single target vessel; The convolutional neural network module in the fusion neural network recognition model is used to extract local dwell features, speed changes, and heading changes from the ship trajectory data. The deep learning model network structure in the fusion neural network recognition model is used to capture the global behavioral dependency of the trajectory segment in the time dimension. The global behavioral dependency is used to determine the dependency of a single ship trajectory segment on a positive or negative sample of a single ship trajectory within a time window. The global representation vector is used as the overall semantic representation of the single ship trajectory segment, and a fully connected classifier is used to output whether the single ship to be identified has engaged in docking behavior.
2. The method for recognizing leaning behavior according to claim 1, characterized in that, The determination of the docking behavior trajectory segment corresponding to the docking behavior based on the first travel trajectory and the first trajectory data of the two vessels includes: Determine the first and second vessels in the dual vessels; Obtain historical data on vessel berthing; Determine the distance threshold for berthing behavior based on the aforementioned historical ship berthing data; Determine multiple trajectory segments of the two vessels; Determine the relative distance between the first vessel and the second vessel; When the relative distance is less than the docking distance threshold and the first trajectory data of the first vessel and the first trajectory data of the second vessel change synchronously, it is determined that there is a docking behavior between the first vessel and the second vessel. At least one of the trajectory segments corresponding to the "leaning" behavior is identified as a "leaning" behavior trajectory segment.
3. The method for recognizing leaning behavior according to claim 2, characterized in that, The step of determining the negative sample dataset of a single ship's trajectory based on the normal navigation trajectory segment includes: Determine multiple normal navigation trajectory segments of the first vessel and the second vessel; Determine the second trajectory data corresponding to the normal navigation trajectory segment; Determine the time-series vector sequence based on the second trajectory data corresponding to the normal navigation trajectory segment; The negative sample dataset of a single ship trajectory is determined based on multiple time-series vector sequences.
4. The method for recognizing leaning behavior according to claim 1, characterized in that, Determining the trajectory segment corresponding to the target trajectory data includes: The target trajectory data is cleaned to remove abnormal data points, and the cleaned target trajectory data is determined. The target trajectory data includes multiple data points. The target vessel's target trajectory is determined based on the cleaned target trajectory data. The target driving trajectory is segmented to determine multiple trajectory segments that include complete speed change cycles, and the data points within the trajectory segments have the same time interval.
5. The method for recognizing leaning behavior according to any one of claims 1 to 4, characterized in that, The trajectory data includes the first trajectory data, the second trajectory data, and the target trajectory data; The data fields corresponding to the trajectory data include: timestamp, longitude, latitude, speed, heading, time interval, and behavior rhythm.
6. A device for reclining behavior recognition, characterized in that, include: The positive sample module is used to acquire the first travel trajectory of the two ships and the corresponding first trajectory data. Based on the first travel trajectory and the first trajectory data of the two vessels, the docking behavior trajectory segment corresponding to the docking behavior is determined; Determine the positive sample dataset of single-ship trajectory based on the aforementioned docking behavior trajectory segments; The negative sample module is used to acquire the second driving trajectory of the two vessels and the corresponding second trajectory data. Based on the second travel trajectory and second trajectory data of the two vessels, the normal navigation trajectory segment corresponding to the normal travel of the vessels is determined; Determine the negative sample dataset of a single ship's trajectory based on the normal navigation trajectory segment; The model training module is used to train a fusion neural network recognition model based on the single ship trajectory positive sample dataset and the single ship trajectory negative sample dataset. The fusion neural network recognition model is configured to take the target trajectory data of a single target ship as input. The data acquisition module is used to acquire the target trajectory data of the single target vessel to be identified; The segment determination module is used to determine the trajectory segment corresponding to the target trajectory data; The docking identification module is used to input the target trajectory data into the fusion neural network identification model to determine the docking behavior status corresponding to at least one trajectory segment of the single target vessel; The convolutional neural network module in the fusion neural network recognition model is used to extract local dwell features, speed changes, and heading changes from the ship trajectory data. The deep learning model network structure in the fusion neural network recognition model is used to capture the global behavioral dependency of the trajectory segment in the time dimension. The global behavioral dependency is used to determine the dependency of a single ship trajectory segment on a positive or negative sample of a single ship trajectory within a time window. The global representation vector is used as the overall semantic representation of the single ship trajectory segment, and a fully connected classifier is used to output whether the single ship to be identified has engaged in docking behavior.
7. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the leaning behavior recognition method as described in any one of claims 1 to 5.
8. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the leaning behavior recognition method as described in any one of claims 1 to 5.
9. A chip, characterized in that, The chip includes a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the steps of the leaning behavior recognition method as described in any one of claims 1 to 5.