Unsupervised ship stay identification method and system based on multi-parameter constraint

Abstract

The invention provides an unsupervised ship stay identification method and system based on multi-parameter constraint, and the method comprises the steps: carrying out the recognition of a jump point for the positioning data of an AIS system in a set range of a port entrance, and dividing the trajectory of each ship into different trajectory segments according to the recognized jump point; and for each track section, identifying whether each track point is a stay point or not in sequence, and classifying anchoring and parking for the identified stay points. The method has the advantages that track jump caused by environmental factors can be effectively recognized, and the defects of a traditional single threshold value method are overcome; the method of directly adopting a single position judgment method is avoided, and more accurate stay behavior recognition is provided; the fuzzy boundary between the stay behavior and the navigation behavior is effectively solved, and the stability and precision of stay behavior recognition are improved.

Description

Technical Field

[0001] The present application belongs to the technical field of ship identification, and specifically relates to an unsupervised ship stop identification method and system based on multi-parameter constraints. Background Art

[0002] Maritime transport plays a vital role in international trade, with more than 90% of global trade carried by sea. With the continuous expansion of maritime trade, the increase in the number, types and activities of ships, especially ships as the main carriers of maritime transport, has led to the ocean becoming an increasingly crowded waterway, and the shipping dynamics have become increasingly complex. This growth has led to increasing risks in navigation safety, and traditional port monitoring and management methods are facing unprecedented challenges. In response to this situation, more and more ports are being developed into digital and intelligent hubs, with the goal of achieving more intelligent, green, efficient and safe operations.

[0003] The identification of ship stop behavior is particularly important in port management, which can help with port entry and exit statistics, cargo loading and unloading activity monitoring, port location detection, and the construction of maritime transportation networks between ports. The ship's stop behavior can generally be divided into two categories: anchorage and berthing. Due to the limited processing capacity of the port, ships often need to anchor in a designated area, wait for the service to be completed, and then berth according to the port scheduling to carry out port-related operations such as cargo handling and passenger services. AIS (Automatic Identification System) data records various information about the ship, including location and attributes, but the navigation status is often manually entered, which leads to frequent errors and delays. In addition, due to the influence of external environmental factors such as wind and tide, the ship will experience irregular fluctuations during berthing or anchoring, making behavior identification more difficult.

[0004] At present, the research methods for single-ship behavior recognition can be mainly divided into three categories: methods based on statistical analysis and navigation feature constraints, methods based on unsupervised clustering of trajectory points, and methods based on supervised deep learning. Traditional statistical analysis methods usually rely on the spatiotemporal dynamics of maritime traffic for modeling. Although these methods can effectively identify hot spots in some situations, their real-time and accuracy are often limited. With the rise of deep learning, more and more research has begun to turn to the use of deep learning models for behavior recognition. For example, models based on convolutional neural networks (CNNs) can identify different forms of behavior by mapping trajectories through grids; graph neural networks (GNNs) identify behavior patterns by treating ship positions as graph nodes and performing node classification; in addition, some studies have also tried to use other machine learning methods, such as random forests and transformers, and achieved good results.

[0005] Although deep learning methods can significantly improve classification accuracy, they usually require a large amount of labeled data and training, and have limited transferability to different sea areas. In order to meet these challenges, more and more researchers have begun to focus on how to combine different parameter constraints, especially under unsupervised conditions, to optimize the recognition algorithm of stop behavior. The advantage of unsupervised methods is that they do not rely on manually labeled training data, can be efficiently processed in real-time data streams, and can adapt to more complex navigation environments and changing port operating conditions. Summary of the invention

[0006] In order to overcome the defects of the prior art, this application proposes an unsupervised ship stop identification method based on multi-parameter constraints, including:

[0007] Step 1: Identify the jump points of the positioning data of the AIS system within the set range of the port entrance, and divide the trajectory of each ship into different trajectory segments according to the identified jump points;

[0008] Step 2: For each track segment, identify whether each track point is a stop point or not, and classify the identified stop points into anchoring and docking;

[0009] The step 2 comprises:

[0010] Step 2-1: In a trajectory segment, the adjacent trajectory points are regarded as vectors, and whether the current point is the ship's stop point is identified based on the cumulative sum of the dot products of the continuous vectors;

[0011] Step 2-2: Classify the stop point as anchoring or docking according to the swing amplitude of the stop segment where the stop point is located.

[0012] As an improvement of the above method, the process of identifying the jump point includes:

[0013] If any of the time difference, distance difference or speed difference between a track point and an adjacent track point exceeds a set threshold, the track point is identified as a jump point.

[0014] As an improvement of the above method, the process of identifying the jump point further includes confirming the identified jump point:

[0015] For each identified jump point, the sum of distances between all point pairs within the previously set jump point range is calculated. If the distance between the point and the previous trajectory point exceeds half of the sum of distances, the point is confirmed to be a true jump point.

[0016] As an improvement of the above method, step 2-1 includes:

[0017] In a trajectory segment, for each trajectory point, taking the point as the center, extracting a first set number of trajectory points before and after, and calculating the cumulative sum of the dot products of the extracted points;

[0018] When the cumulative sum of the dot products is less than a first set dot product sum threshold, the point is set as a stay point.

[0019] As an improvement to the above method, for the trajectory points at the beginning and end of the trajectory segment, if the number of trajectory points before or after the trajectory point is less than half of the first set number, the setting of whether the trajectory point is a stop point is consistent with its closest identified point.

[0020] As an improvement of the above method, after step 2-1, the method further includes:

[0021] Step 2-1A: If the current trajectory point is set as a stop point, the point is stored in a set of set elements. If the set is full, the first stop point stored is deleted.

[0022] Calculate the mean and maximum of the positions of all the stay points in the set;

[0023] For the current trajectory point, if the difference between the point and the position mean is less than the maximum position minus the position mean, the point is confirmed as a stop point.

[0024] As an improvement of the above method, after step 2-1A, the method further includes:

[0025] Step 2-1B: For the current track point, take a second set number of track points as the data to be reviewed;

[0026] If the number of stay points in the review data is greater than the set multiple of the number of non-stay points, all points between the first stay point and the last stay point will be finally confirmed as stay points;

[0027] If the review data meets both of the following conditions, all review data will be set as travel points:

[0028] Condition 1: The first and last trajectory points are both driving points;

[0029] Condition 2: The dot product sum of all points between the first and last set stay points in the review data is greater than the second set dot product sum threshold, or the number of stay points set in the review data is less than the set proportion of the total number of review data.

[0030] As an improvement of the above method, the setting multiple is 2; the setting ratio is 1 / 10.

[0031] As an improvement of the above method, step 2-2 includes:

[0032] If the current trajectory point is a stop point, take the point as the center, extract the first set number of trajectory points before and after it as the stop segment of the stop point, and calculate the cumulative sum of the dot products of all the points in the stop segment;

[0033] If the cumulative sum of the dot products is greater than a second set dot product sum threshold, the point is determined to be an anchor, otherwise the point is determined to be a stop.

[0034] The present application also provides an unsupervised ship stop identification system based on multi-parameter constraints, which is implemented based on the above method. The system includes:

[0035] The jump point identification module is used to identify the jump points of the positioning data of the AIS system within the set range of the port entrance, and divide the trajectory of each ship into different trajectory segments according to the identified jump points;

[0036] The stop point recognition and classification module is used to regard adjacent trajectory points as vectors in each trajectory segment, and identify whether the current point is the ship's stop point based on the cumulative sum of the dot products of continuous vectors; and classify the stop point as anchoring or docking based on the swing amplitude of the stop segment where the stop point is located.

[0037] Compared with the prior art, the advantages of this application are:

[0038] 1. The method of the present application is a method for jump point identification based on multiple thresholds. The present application proposes a jump point identification method that combines time, distance and speed thresholds to identify jump points by evaluating the differences between adjacent trajectory points (such as time difference, distance difference and speed difference). In addition, a time window method is used to further improve the accuracy of jump point identification by calculating the sum of distances between point pairs within a certain window. This method can effectively identify trajectory jumps caused by environmental factors (such as ocean currents, wind direction, etc.), making up for the shortcomings of the traditional single threshold method.

[0039] 2. The method of the present application is a dot product-based ship stop identification method. When identifying ship stop behavior, the present application proposes a recognition method based on trajectory dot product. By calculating the dot product of adjacent trajectory points and accumulating them, a threshold is combined to determine whether the current trajectory point represents a stop state. This method can handle the random swing of the ship caused by environmental factors, avoids the direct use of a single position judgment method, and provides more accurate stop behavior recognition.

[0040] 3. The method of the present application has a continuity correction mechanism for stop behavior. In order to further improve the accuracy of stop behavior recognition, the present application proposes a correction mechanism based on the review of surrounding points. The misclassification of stop and navigation behaviors is corrected by storing previously identified stop points and combining the recognition results of surrounding trajectory points. The continuity correction mechanism can effectively resolve the fuzzy boundary between stop behavior and navigation behavior, and improve the stability and accuracy of stop behavior recognition.

[0041] These three advantages together build a more accurate and robust method for ship behavior identification, especially suitable for complex marine environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Shown is a schematic diagram of the principle of the unsupervised ship stop identification method based on multi-parameter constraints. DETAILED DESCRIPTION

[0043] The technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0044] This application proposes an unsupervised ship stop identification method and system based on multi-parameter constraints. Through reasonable parameter design, combined with ship trajectory characteristics, environmental conditions and port operation data, efficient and accurate identification of ship stop behavior is achieved. This method can fully explore the implicit behavioral laws in the process of ship stop without a large amount of labeled data, solve the challenges faced by existing technologies in processing large-scale, real-time data, and provide important support for port management and shipping safety.

[0045] The purpose of this application is to solve the accuracy and real-time problems of ship stop behavior recognition in current port management. With the continuous growth of maritime transportation, ships enter and leave ports frequently, and the traffic volume in ports increases year by year, resulting in an increase in collision risks and increased management difficulties. Traditional behavior recognition methods often rely on manually input navigation status or require a large amount of labeled data support, which is prone to errors and difficult to cope with complex marine environments. Existing statistical methods and recognition models based on supervised learning usually have problems such as poor real-time performance and weak transferability, which make it difficult to meet the needs of dynamic port environments.

[0046] This application provides a method for identifying stop behaviors without manual annotation by designing multi-parameter constraints and combining multiple features such as ship speed and position changes. This method is based on the idea of ​​unsupervised learning and can automatically identify the anchorage and berthing behaviors of ships in large-scale AIS data, effectively reducing manual intervention while ensuring recognition accuracy. In addition, this method vectorizes the trajectory, fully considers the displacement fluctuations caused by external factors such as wind, waves, and tides during the ship's stay, and filters non-stop behaviors through multi-parameter constraints, thereby enhancing the robustness of recognition. This method can quickly analyze ship trajectories in the case of large data volumes and real-time data streams, and is suitable for complex maritime traffic environments and changing port operation needs.

[0047] The implementation of this application method will contribute to the intelligent management of ports, improve the efficiency of traffic flow monitoring, reduce the risk of collision, and provide support for application scenarios such as port entry and exit statistics, cargo loading and unloading management, and shipping network construction. At the same time, the unsupervised algorithm improves the adaptability and scalability of the system, effectively responds to the application needs of different ports and environments, and promotes the intelligent, precise and efficient development of port management.

[0048] Example 1

[0049] like Figure 1 As shown, the present application provides an unsupervised ship stay identification method based on multi-parameter constraints, which is used to identify the stay behavior of ships near the port, mainly including the identification of different stay behaviors, berthing and anchoring. At the same time, it identifies the jump points caused by data quality problems.

[0050] Before introducing the detailed method flow of this application, we first conduct a modeling analysis of the problem. The single ship behavior recognition problem is a classic classification task, which can be regarded as a learning function. To estimate the most likely behavior of a ship given its basic attributes and trajectory characteristics. Therefore, the classification formula in this application can be expressed as:

[0051]

[0052] in, It is a single ship behavior recognition method. (V,T) defines the dynamic trajectory characteristics and static attribute characteristics of the ship. is the behavior label predicted by this method, where V = {type, Legth, width}, including the ship type, length and width of the ship. Where x = (t, λ, φ, v, θ), which are timestamp, longitude and latitude, and speed, respectively. For specific meanings, please refer to Table 1. i is the index of different time points. is a parameter in the model. At the same time, Figure 1As shown, the method proposed in this application does not require advance labeling of data, which reduces a lot of manpower labeling costs and computing power costs, and can realize real-time ship trajectory behavior recognition.

[0053] Table 1 Variables used in this application and their meanings

[0054]

[0055] Near the port entrance, the positioning data of the AIS system has certain errors or omissions, and data transmission may be interrupted, resulting in no position updates in certain periods of time, or after the data is restored, there are jump differences between two consecutive points, which are called jump points. These problems are mainly due to the data compression processing performed by NOAA (National Oceanic and Atmospheric Administration) before public release. Since the changes in the ship's stay and forward status are usually gradual and not sudden, if the trajectory is segmented and identified directly based on all points, without segmenting the trajectory according to the jump points, the continuity and gradualness of the state change will be ignored, which may lead to misjudgment or omission, and cannot accurately reflect the true dynamics of the ship's behavior. Therefore, we set more refined thresholds to better identify jump points in various aspects.

[0056] Jump point identification method: By applying thresholds of time, distance and speed, we identify significant jumps (VesP) between consecutive trajectory points. If the time difference (∈ time ), distance difference (∈ dis ) or speed difference (∈ velocity ) exceeds the predefined threshold, the point is classified as a jump point (see Algorithm 1 in Table 2, steps 2-4). In order to further improve the accuracy of jump point identification, we introduce a method based on time window. For each trajectory point, we calculate the time window before ω jump The sum of distances between all pairs of points within the range of the jump point. If the distance between the current point and the previous point exceeds half of the sum of these distances, the current point is marked as a jump point (see Algorithm 1 in Table 2, steps 6-9). This additional check ensures that sudden and significant changes in the trajectory are captured, even if these changes may not be captured by a single threshold.

[0057] Table 2 Jump point recognition and trajectory segmentation algorithm

[0058]

[0059] Once the jump points are identified, we divide each ship’s trajectory into different segments for further analysis. These trajectory segments are generated by dividing the trajectory at each identified jump point (see Algorithm 1 in Table 2, steps 12-20). This segmentation is crucial for subsequent processing, such as identifying ship stop behaviors.

[0060] Table 3. Stay point identification algorithm

[0061]

[0062] We assume that the ship trajectory is continuous, that is, there is no long period of data loss or the ship suddenly jumps across land from one location to another. Therefore, as described above, we segment the trajectory according to the identified jump points. For ship behavior recognition, this application mainly focuses on identifying stop features and distinguishing between docking and anchoring behaviors. When a ship is docked or anchored, it is not fixed to a specific longitude and latitude position, but is affected by factors such as ocean currents, wind direction, tides, and instrument errors, resulting in random and irregular movement of the ship's trajectory. During the docking process, the ship is connected to the port by a short cable, causing the ship to swing randomly near a specific location. During the anchoring process, the ship is fixed to the seabed by a long chain, and the ship's trajectory swings within a circular area.

[0063] For each trajectory segment divided by a jump point, we determine whether each trajectory point in the trajectory segment is a stop point, and then classify the determined stop points as anchoring or docking.

[0064] The process of identifying and classifying a trajectory point is as follows:

[0065] We treat adjacent trajectory points as vectors, normalize them, and calculate the dot product of continuous vectors (see Algorithm 2, steps 2-3 in Table 3). Considering that the ship will swing within a certain range when it stops, the cumulative sum of the dot products should be close to zero over a period of time. In order to determine whether the ship is stopping, the front and rear trajectory points are extracted with the currently identified trajectory point as the center. points, and calculate the cumulative sum of the dot products in the entire window. Then, use the threshold ξ judge To determine whether the current point represents a stop (see Algorithm 2 in Table 3, steps 2-5), the cumulative sum of the dot products is less than the threshold ξ judge The trajectory points at the beginning and end of the trajectory segment are identified as stop points. Points within range will have their status consistent with the closest identified point. This approach reduces false positives. For example, if a vessel is turning or changing direction, a single point may appear to be reversed, but the overall pattern of surrounding points will be correctly identified as non-stationary.

[0066] In addition, unlike the docking behavior, when the ship is docked, it swings slightly around a fixed position, while when it is anchored, the swing amplitude of the ship is larger and the period is longer. judge Setting it too long may reduce the accuracy and algorithm efficiency. To address this issue, we introduce an additional speed constraint.

[0067] The ship's stop behavior is a continuous process; therefore, after the initial identification, we proceed to the review step based on the surrounding point identification results. First, we store the previously identified stop points in a save , and further refine the identification of trajectory points based on their positional relationship with the currently identified stop point (see Algorithm 2 in Table 3, steps 7-10). Specifically, when a trajectory is judged to be a stop point, it is added to the queue. The length of this queue is fixed. If the amount of data exceeds the set length, the first data stored will be deleted. When a point is to be reviewed, the mean and maximum values ​​of the data positions in the queue are first calculated. If the difference between the current point to be identified and the position mean is less than the maximum value of the position minus the mean value of the position, the point is a stop point. Secondly, considering the continuity of stop and navigation behaviors, we correct the misclassification of stop and navigation behaviors by referring to the recognition results of surrounding points (see Algorithm 2 in Table 3, steps 14-19). Here we start with the currently judged node and take ω forward. check The data are taken as the data to be reviewed. If the number of stay points in the review data is greater than twice the number of non-stay points, the labels of all points between the first stay point and the last stay point are set to stay. If the review data meets two conditions, the identification labels of all review data are set to travel. Condition 1 is that the first point and the last point are both travel points, and condition 2 is that the sum of the points between the first and the last points initially identified as stop points in the review data is greater than ξ check , or the number of points initially identified as stays in the review data is less than one-tenth of the total number of review data. Finally, we classify the trajectory points that have been confirmed to belong to stay behaviors. If the swing amplitude of the stay segment is large, that is, the cumulative sum of the dot products exceeds ξ check ——, it is classified as anchored; otherwise, it is classified as docked (see Algorithm 2 in Table 3, steps 25-27).

[0068] The following uses a mature public dataset with complete stop trajectory annotation (https: / / github.com / destiny1103 / DT-GNN) to illustrate the accuracy of the method provided by this application. This dataset is an expert-annotated stop behavior recognition result dataset around New York Harbor, with labels including: docked (78.1%), moored (4.8%), and sailing (17.0%). This dataset has clear labels and clean data. The preparation and division of the dataset provides a solid foundation for testing.

[0069] This application uses the following key indicators to evaluate the performance of ship behavior recognition: accuracy, precision, recall and F1 value. The calculation formula for each indicator is as follows:

[0070] Accuracy: measures the proportion of correct predictions to the total number of predictions. The calculation formula is:

[0071]

[0072] Among them, T P is the number of true positives, T N is the number of true negatives, F P is the number of false positives, F N is the number of false negatives.

[0073] Precision: Quantifies the accuracy of positive predictions, reflecting the proportion of correctly predicted positive instances. The calculation formula is:

[0074]

[0075] Recall: It evaluates the model's ability to identify all positive instances, indicating the proportion of correctly identified positive instances to all actual positive instances. The calculation formula is:

[0076]

[0077] F1-score: The harmonic mean of precision and recall, providing a balance between the two. It is particularly suitable for situations where the category distribution is uneven. The calculation formula is:

[0078]

[0079] These indicators comprehensively evaluate the performance of the model in identifying different ship stop behaviors.

[0080] Comparison method introduction:

[0081] We compared multiple models, including rule-based constraints, clustering methods, and several graph neural networks. The specific methods are as follows:

[0082] Rule-based constraint method (RBC): This method identifies stationary trajectories by defining rules and constraints based on the velocity characteristics of trajectory points. We implement this method by directly setting the velocity threshold.

[0083] Density-based clustering method: This method identifies stationary trajectories by clustering trajectory points with similar spatial locations. We use the DBSCAN algorithm to implement this method.

[0084] Graph Convolutional Network (GCN): Graph Convolutional Network extends the convolution operation to the graph structure, which can capture the complex dependencies between nodes and extract global structural information.

[0085] Graph Attention Network (GAT): Graph Attention Network introduces an attention mechanism to dynamically assign different importance weights to nodes in the graph, allowing the model to pay more attention to information-rich neighboring nodes.

[0086] ChebNet: ChebNet uses Chebyshev polynomials as filters in graph convolution to enhance the expressive power of graph networks.

[0087] The geographical layout of a port may be complex, with multiple intersections. Accurately identifying jump points based on distance alone is challenging: if the threshold is set too short, fast-moving ships may be mistakenly identified as jump points; if it is set too long, jump points at turns may not be detected. Although jump point identification at turns does not affect the behavior identification of a single ship, it may cause problems when aligning the trajectories of multiple ships in time, as the interpolated trajectory may be mapped onto land. Since ships in ports usually move at lower speeds, we set a lower speed threshold. Other thresholds were also adjusted according to specific characteristics, as shown in Table 4:

[0088] Table 4 Jump point identification related parameters

[0089]

[0090] In Table 5, the window-based parameter ω judge ,ω stop and ω check It can be scaled according to different data sets. The original data unit is meter per second (m / s). During parameter tuning, ν check is set to 0.2m / s, which translates to 0.72km / h. Some datasets are sparse, with fewer stops, and require a smaller window size. For the initial parameter ξ used to determine the stop judge and ν judge , should be adjusted according to the characteristics of the ship. For example, large cargo ships usually move slowly when entering and leaving the port for safety reasons, while small ships such as sailboats and passenger ships usually move faster. For faster ships, these parameters should be set to smaller values ​​to more effectively detect the anchorage period. As for the calibration parameter ξ check , usually only slightly larger than ξ judge However, for sparse trajectory data, a larger ξ is required. check If the trajectories are denser, a larger ξ check Turning points may be mistakenly identified as rest periods.

[0091] Table 5 Parameter settings in the comparison experiment of stay point recognition accuracy.

[0092]

[0093] Comparative analysis of recognition accuracy:

[0094] The performance comparison in Table 6 demonstrates the effectiveness of the method proposed in this application, which achieved a high accuracy of 98.738% and an F1 score of 98.868% when identifying stationary trajectories, indicating that it has balanced precision and recall when accurately identifying stationary trajectories. In contrast, the rule-based constraint (RBC) method has an accuracy of 96.871%, but its precision is lower, indicating that it has a tendency to over-predict stationary points. The DBSCAN method has an accuracy of 81.224% and a low recall, which may be due to its over-sensitivity to parameter settings and difficulties in dealing with trajectories of different densities. The results of RBC and DBSCAN represent the most accurate results obtained under different parameter settings. For the four graph neural networks, 80% of the data were randomly selected for training, and the table shows the accuracy of each model on the entire dataset.

[0095] Table 6 Performance comparison of different methods.

[0096]

[0097] Three graph neural network methods (GAT, GCN and ChebNet) show good accuracy, among which the method of this application leads with an accuracy of 98.738%. However, it should be noted that these graph neural network models are trained on 80% of the data, so they have certain data advantages compared to the method proposed in this application, because the method proposed in this application is evaluated based on unseen data. The accuracy of most graph neural network methods is slightly lower than that of the method of this application. The main reason is that the changes in the behavior state of individual ships are gradual, and the design of the GNN structure limits its ability to capture long-range dependencies. In addition, the comparative method in this application relies on the graph structure of spatial modeling, while the behavior state of the ship does not completely depend on the spatial position. The method proposed in this application not only identifies the behavior of a single ship based on the change of the ship's own state, but also further improves the accuracy by correcting the state of the current point and combining it with the state of the previous and next trajectory points. In addition, graph-based methods rely on pre-built graph structures, and these graph structures cannot be updated in real time, which limits their application in real-time scenarios. Although graph neural network-based methods have achieved high accuracy, the method proposed in this application has higher accuracy, stronger real-time capabilities, and can adapt to dynamic data, making it more practical in actual marine applications.

[0098] Table 7 Accuracy of DBSCAN and RBC under different parameters

[0099]

[0100] Table 7 shows the accuracy of DBSCAN and RBC under different parameter settings. Although the RBC method has the advantage of fast real-time decision-making, it is highly sensitive to a single parameter. Its accuracy varies with the ground speed (SOG) threshold, reaching an accuracy of 96.870% at SOG = 3.6, but significant adjustments are required in different scenarios. Similarly, DBSCAN reached its highest accuracy of 81.224% at Eps = 0.0003 and MinPts = 300. However, although DBSCAN performs more robustly under different parameter settings, its principle of matching training data with real-time data has an important impact on the migration of the model in different sea areas. In comparison, although the method proposed in this application also requires parameter tuning, it includes decision parameters and calibration steps, thereby improving its adaptability and accuracy under different conditions.

[0101] The method provided in this application proposes a framework based on single-ship behavior recognition for the problem of ship behavior recognition, and has achieved remarkable technical effects in terms of accuracy and real-time performance. Compared with the traditional rule-based constraint method (RBC) and density-based clustering methods (such as DBSCAN), the method proposed in this application can balance precision and recall while being highly accurate, and accurately identify the static trajectory of the ship. In particular, the method can significantly improve the accuracy of behavior recognition by correcting the recognition result of the current point and combining the state information of the previous and next trajectory points, thus solving the shortcomings of traditional methods in complex environments. In addition, compared with the graph neural network (GNN) method, although the accuracy of this method is slightly lower, its advantages in real-time and dynamic data adaptability make it more suitable for practical shipping applications. This method not only has strong adaptability in complex scenarios such as ports, but also can effectively handle changes in different sea areas and ship types, ensuring accuracy and robustness under different conditions.

[0102] Example 2

[0103] The present application also provides an unsupervised ship stop identification system based on multi-parameter constraints, which is implemented based on the above method. The system includes:

[0104] The jump point identification module is used to identify the jump points of the positioning data of the AIS system within the set range of the port entrance, and divide the trajectory of each ship into different trajectory segments according to the identified jump points;

[0105] The stop point recognition and classification module is used to regard adjacent trajectory points as vectors in each trajectory segment, and identify whether the current point is the ship's stop point based on the cumulative sum of the dot products of continuous vectors; and classify the stop point as anchoring or docking based on the swing amplitude of the stop segment where the stop point is located.

[0106] The present application may also provide a computer device, comprising: at least one processor, a memory, at least one network interface and a user interface. The various components in the device are coupled together through a bus system. It is understood that the bus system is used to achieve connection and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus and a status signal bus.

[0107] The user interface may include a display, a keyboard or a pointing device, such as a mouse, a trackball, a touch pad or a touch screen.

[0108] It is understood that the memory in the embodiments disclosed in the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as 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 random access memory (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.

[0109] In some embodiments, the memory stores the following elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and applications.

[0110] The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application includes various application programs, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present disclosure can be included in the application.

[0111] In the above embodiment, the processor may also call a program or instruction stored in the memory, specifically, a program or instruction stored in an application program, and is used to:

[0112] Execute the steps of the above method.

[0113] The above method can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The above-disclosed methods, steps and logic block diagrams can be implemented or executed. The general processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the above-disclosed method can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined to execute. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0114] It is understood that the embodiments described in the present application can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.

[0115] For software implementation, the technology of the present application can be implemented by executing the functional modules (such as procedures, functions, etc.) of the present application. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0116] The present application may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application is described in detail with reference to the embodiments, a person skilled in the art should understand that any modification or equivalent replacement of the technical solution of the present application does not depart from the spirit and scope of the technical solution of the present application and should be included in the scope of the claims of the present application.

Claims

1. An unsupervised ship stop identification method based on multi-parameter constraints, comprising: Step 1: Identify the jump points of the positioning data of the AIS system within the set range of the port entrance, and divide the trajectory of each ship into different trajectory segments according to the identified jump points; Step 2: For each track segment, identify whether each track point is a stop point or not, and classify the identified stop points into anchoring and docking; The step 2 comprises: Step 2-1: In a trajectory segment, the adjacent trajectory points are regarded as vectors, and whether the current point is the ship's stop point is identified based on the cumulative sum of the dot products of the continuous vectors; Step 2-2: Classify the stop point as anchoring or docking according to the swing amplitude of the stop segment where the stop point is located.

2. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 1 is characterized in that: The process of identifying the jump point includes: If any of the time difference, distance difference or speed difference between a track point and an adjacent track point exceeds a set threshold, the track point is identified as a jump point.

3. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 2 is characterized in that: The process of identifying the jump point also includes confirming the identified jump point: For each identified jump point, the sum of distances between all point pairs within the previously set jump point range is calculated. If the distance between the point and the previous trajectory point exceeds half of the sum of distances, the point is confirmed to be a true jump point.

4. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 1 is characterized in that: The step 2-1 comprises: In a trajectory segment, for each trajectory point, taking the point as the center, extracting a first set number of trajectory points before and after, and calculating the cumulative sum of the dot products of the extracted points; When the cumulative sum of the dot products is less than a first set dot product sum threshold, the point is set as a stay point.

5. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 4 is characterized in that: For the trajectory points at the beginning and end of the trajectory segment, if the number of trajectory points before or after the trajectory point is less than half of the first set number, the setting of whether the trajectory point is a stop point is consistent with the closest identified point.

6. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 4 is characterized in that: After step 2-1, the method further includes: Step 2-1A: If the current trajectory point is set as a stop point, the point is stored in a set of set elements. If the set is full, the first stop point stored is deleted. Calculate the mean and maximum of the positions of all the stay points in the set; For the current trajectory point, if the difference between the point and the position mean is less than the maximum position minus the position mean, the point is confirmed as a stop point.

7. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 6 is characterized in that: After step 2-1A, the method further includes: Step 2-1B: For the current track point, take a second set number of track points as the data to be reviewed; If the number of stay points in the review data is greater than the set multiple of the number of non-stay points, all points between the first stay point and the last stay point will be finally confirmed as stay points; If the review data meets both of the following conditions, all review data will be set as travel points: Condition 1: The first and last trajectory points are both driving points; Condition 2: The dot product sum of all points between the first and last set stay points in the review data is greater than the second set dot product sum threshold, or the number of stay points set in the review data is less than the set proportion of the total number of review data.

8. The unsupervised ship stop identification method based on multi-parameter constraints according to claim 7 is characterized in that: The setting multiple is 2; the setting ratio is 1 / 10.

9. The unsupervised ship stop identification method based on multi-parameter constraints according to any one of claims 1 to 8, characterized in that: The step 2-2 comprises: If the current trajectory point is a stop point, take the point as the center, extract the first set number of trajectory points before and after it as the stop segment of the stop point, and calculate the cumulative sum of the dot products of all the points in the stop segment; If the cumulative sum of the dot products is greater than a second set dot product sum threshold, the point is determined to be an anchor, otherwise the point is determined to be a stop.

10. An unsupervised ship stop identification system based on multi-parameter constraints, implemented based on any of the methods described in claims 1-9, characterized in that: The system comprises: A jump point identification module is used to identify jump points of the positioning data of the AIS system within a set range of the port entrance, and divide the track of each ship into different track segments according to the identified jump points; and The stop point recognition and classification module is used to regard adjacent trajectory points as vectors in each trajectory segment, and identify whether the current point is the ship's stop point based on the cumulative sum of the dot products of continuous vectors; and classify the stop point as anchoring or docking based on the swing amplitude of the stop segment where the stop point is located.

Images