Methods and systems for identifying ship spoofed tracks, electronic devices, and storage media.

Abstract

This invention relates to the field of shipbuilding technology, and proposes a method and system for identifying forged ship trajectories, an electronic device, and a storage medium. The method for identifying forged ship trajectories includes: acquiring AIS data and a graph autoencoder model; extracting the ship trajectory and information of each ship from the AIS data; constructing a cooperative behavior graph among multiple ships based on the ship trajectory and information of each ship; determining outliers in each ship trajectory based on the cooperative behavior graph and the graph autoencoder model; and identifying forged trajectories based on the outliers in each ship trajectory. This invention identifies forged trajectories by constructing a cooperative behavior graph among ships and automatically learning the cooperative patterns of normal ship groups using a graph autoencoder model. This eliminates the need for manual labeling, enabling automatic detection of forged trajectories even in large-scale, unlabeled AIS data.

Description

Technical Field

[0001] This invention relates to the field of marine technology, and more specifically, to a method and system for identifying spoofed ship tracks, an electronic device, and a storage medium. Background Technology

[0002] Existing methods for detecting ship spoofed trajectories mainly fall into two categories: rule-based thresholding and supervised learning. Rule-based thresholding judges anomalies based on ship indicators such as speed, heading, and time intervals, but lacks modeling of coordinated ship behavior, making it prone to false positives or false negatives. Supervised learning trains a classifier based on labeled "normal / spoofed" samples, but its limitations include high labeling costs, extremely scarce spoofed samples, poor model generalization, and a significant decrease in accuracy after changes in sea area or time. Therefore, in the context of large-scale, unlabeled AIS data, these techniques struggle to automatically detect abnormal trajectories. Summary of the Invention

[0003] The present invention aims to solve the problem of difficulty in automatically detecting abnormal trajectories in existing or related technologies.

[0004] Therefore, the first aspect of the present invention proposes a method for identifying ship spoofing tracks.

[0005] A second aspect of the present invention provides a system for identifying ship spoofing tracks.

[0006] A third aspect of the present invention provides an electronic device.

[0007] A fourth aspect of the present invention provides a storage medium.

[0008] In view of this, according to a first aspect of the present invention, a method for identifying ship spoofing trajectories is proposed, wherein the method includes: acquiring AIS data and a graph autoencoder model; extracting the ship trajectory and ship information of each ship from the AIS data; constructing a cooperative behavior graph among multiple ships based on the ship trajectory and ship information of each ship; determining outliers in each ship trajectory based on the cooperative behavior graph and the graph autoencoder model; and determining the spoofing trajectory based on the outliers in each ship trajectory.

[0009] The method for identifying spoofed ship trajectories provided by this invention mainly includes: first, acquiring AIS (Automatic Identification System) data containing information about multiple ships and a pre-trained graph autoencoder model. The graph autoencoder model is an unsupervised learning model based on graph neural networks, aiming to learn a low-dimensional representation of the graph through the structure of the encoder and decoder, and reconstruct the adjacency matrix of the graph. Then, extracting the ship trajectory and ship information for each ship from the mixed AIS data. Extracting the ship trajectory for each ship can involve first determining the overall navigation trajectory of each ship, then segmenting the overall navigation trajectory according to a preset time window to obtain trajectory segments for each ship under different time windows, and finally integrating them according to the time windows. In other words, the ship trajectory of each ship mentioned in this invention refers to the ship trajectories of different ships under the same time window. Ship information refers to the static attribute information of the ship, such as ship type, ship name, and ship length. After obtaining the ship trajectory and information of each vessel, a collaborative behavior graph among multiple vessels is constructed based on these parameters. In other words, the behavioral interactions of the vessel group are modeled as a graph structure, containing the collaborative patterns of different vessels during navigation. The constructed collaborative behavior graph is then input into a pre-trained graph autoencoder model to obtain outliers corresponding to each vessel trajectory. The graph autoencoder model can perform unsupervised learning of the collaborative behavior graph, calculating the reconstruction error, i.e., the outlier, of each vessel trajectory through its encoding-decoding mechanism. Finally, based on the outliers corresponding to each vessel trajectory, abnormal or forged trajectories deviating from the normal collaborative pattern are automatically identified; that is, forged trajectories are determined from multiple vessel trajectories based on the outliers of each vessel trajectory. In this invention, because the graph autoencoder model learns the universal "collaborative behavior" patterns among vessel groups, rather than local anomaly patterns specific to a particular sea area, it possesses excellent cross-regional and cross-time-time transfer and generalization energy, overcoming the performance degradation problem after scene transformation in related technologies. Furthermore, in this invention, the detection results are derived from the reconstruction error, i.e., the outlier, of the vessel trajectory and the group's collaborative pattern, thus providing a direct explanation of the causes of anomalies. This invention constructs a cooperative behavior map among ships and uses a graph autoencoder model to automatically learn the cooperative patterns of normal ship groups, thereby identifying forged trajectories. It completely eliminates the dependence on hard-to-obtain labeled data for "forged trajectories" and can directly use massive amounts of unlabeled AIS data for detection, solving the core pain points of high labeling costs and scarcity of forged samples in supervised learning methods.

[0010] In some technical solutions, optionally, the step of constructing a cooperative behavior graph among multiple ships based on the ship trajectory and ship information of each ship includes: treating the ship trajectories of multiple ships as multiple nodes; counting the number of times each ship trajectory is received at different base stations; determining the feature vector of each node based on the number of receptions and ship information; calculating the trajectory spatial similarity between any two nodes; determining multiple edges based on the trajectory spatial similarity between any two nodes; calculating the hull difference between two nodes connected to the edge based on the ship information of each ship; using the trajectory spatial similarity and hull difference between two nodes connected to the edge as edge feature vectors; and constructing a cooperative behavior graph based on the nodes, edges, node feature vectors, and edge feature vectors.

[0011] In this technical solution, the steps of constructing a collaborative behavior graph among multiple ships based on the ship trajectory and ship information of each ship include: first, using the ship trajectory of each ship as a node in the collaborative behavior graph; then, counting the number of times each ship's trajectory is received at different base stations, and using the number of receptions as a reception vector reflecting the ship's communication mode; then, determining the feature vector of each node based on the number of receptions and ship information, that is, combining the reception vector with the ship's static attribute information as the node's feature vector; simultaneously, calculating the trajectory spatial similarity between any two nodes, and determining edges based on the trajectory spatial similarity between any two nodes. For example, when the trajectory spatial similarity between any two nodes is greater than a preset similarity, an edge is established between these two nodes, connecting them; if the trajectory spatial similarity between any two nodes is not greater than the preset similarity, no edge is established between these two nodes. By establishing edges between two nodes with trajectory spatial similarity greater than the preset similarity, the collaborative relationship between ships is represented. Then, based on the ship information of each ship, the ship difference between the two nodes connected by the edge is calculated, and then the trajectory spatial similarity and the ship hull difference are used as the feature vector of this edge, i.e., the edge feature vector. In this invention, each edge is assigned a feature vector. This feature vector is composed of the trajectory spatial similarity between the two nodes and the corresponding ship hull differences, thus precisely quantifying the strength and reliability of the cooperative relationship. Ultimately, through these four elements—nodes, edges, node feature vectors, and edge feature vectors—a complete graph capable of deeply expressing the multi-dimensional cooperative behavior of ship groups across space, signals, and attributes is constructed. By introducing the receiving vector and ship hull differences, this invention expands the cooperative relationship from a single spatial similarity to multiple dimensions such as communication behavior matching degree and ship physical attribute compatibility, resulting in a richer information content in the constructed graph and a more accurate description of cooperative behavior.

[0012] In some technical solutions, optionally, the step of calculating the trajectory spatial similarity between any two nodes includes: calculating the Hausdorff distance between any two nodes based on the spatial position of each ship trajectory; and determining the trajectory spatial similarity based on the Hausdorff distance.

[0013] In this technical solution, the step of calculating the trajectory spatial similarity between any two nodes includes: firstly, calculating the Hausdorff distance between any two nodes based on the spatial position of each ship trajectory; and then determining the trajectory spatial similarity based on the Hausdorff distance. In other words, this invention uses the Hausdorff distance as the core algorithm for calculating the trajectory spatial similarity between any two ship trajectories. By calculating the maximum and minimum distances between all points on one ship trajectory and all points on another ship trajectory, it accurately measures the overall matching degree and consistency of the two ship trajectories in the entire spatial morphology, thereby transforming the abstract "spatial similarity" into a computable and stable data metric.

[0014] In some technical solutions, optionally, the graph autoencoder model includes an encoder and a decoder. The step of determining outliers for each ship trajectory based on the cooperative behavior graph and the graph autoencoder model includes: determining the original adjacency matrix corresponding to each node based on the cooperative behavior graph; determining the node embedding vector corresponding to the node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder; determining the reconstructed adjacency matrix corresponding to each node based on the node embedding vector and the decoder; and determining the outliers for each node based on the original adjacency matrix and the reconstructed adjacency matrix.

[0015] In this technical solution, the graph autoencoder model includes an encoder and a decoder. The steps for determining outliers in each ship trajectory based on the cooperative behavior graph and the graph autoencoder model include: First, extracting the original adjacency matrix representing the true connections between nodes from the cooperative behavior graph; then, inputting the information from the cooperative behavior graph—namely, the feature vectors of nodes, the feature vectors of other nodes connected to a node, and the edge feature vectors between two nodes—into the encoder. The encoder maps each node to a low-dimensional node embedding vector containing contextual information about itself and its neighbors (i.e., other nodes connected to the node) through an information aggregation mechanism. Next, the decoder reconstructs the graph's connection relationships using these node embedding vectors through operations such as inner product, generating a reconstructed adjacency matrix. Finally, the difference between the corresponding row (or column) vectors of each node in the original adjacency matrix and the reconstructed adjacency matrix is ​​calculated, and this difference is used as the outlier of that node. This invention, through an "encoding-decoding-comparison" process, transforms the learning of "normal" cooperative behavior patterns into the ability to reconstruct graph structures, automatically identifying nodes that cannot be accurately reconstructed by the model as outliers, thus achieving the detection of forged trajectories without any manual annotation.

[0016] In some technical solutions, optionally, the step of determining the node embedding vector corresponding to the node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder includes: determining the attention weight corresponding to each other node based on the node's feature vector, the feature vectors of other nodes connected to the node, and the edge feature vector between the two nodes; and determining the node embedding vector corresponding to the node based on the attention weight of each other node, the node's feature vector, and the encoder.

[0017] In this technical solution, the step of determining the node embedding vector corresponding to a node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vectors between two nodes, and the encoder includes: First, for the current central node, its own feature vector, the feature vectors of each of its neighboring nodes, and the edge feature vectors connecting them are comprehensively considered to dynamically calculate the attention weight for each neighboring node. Here, a neighboring node refers to a node connected to the central node through an edge, and the central node is any node in the cooperative behavior graph. The attention weight can accurately quantify the importance of different neighboring nodes to the central node. Subsequently, the encoder does not aggregate all neighboring node information equally, but rather performs a weighted summation of the feature vectors of each neighboring node based on the calculated attention weight. Then, the weighted summation feature vector is fused with the central node's feature vector to form a node embedding vector that more accurately reflects the central node's behavior in its specific cooperative context. This invention, by determining the attention weight corresponding to each neighboring node, prevents the model from treating all spatially proximate neighboring nodes equally, instead focusing on those neighboring nodes with highly correlated behaviors and a high probability of cooperative interaction, thereby significantly improving the quality and relevance of information aggregation.

[0018] In some technical solutions, optionally, the step of determining a fake trajectory among multiple ship trajectories based on the outlier of each ship trajectory includes: comparing the outlier of each ship trajectory with a preset threshold; and determining the ship trajectory as a fake trajectory when the outlier of the ship trajectory is greater than the preset threshold.

[0019] In this technical solution, the step of identifying a forged trajectory from multiple ship trajectories based on the outlier of each ship trajectory includes: firstly, comparing the outlier of each ship trajectory with a preset threshold; when the outlier of a ship trajectory exceeds the preset threshold, this ship trajectory is automatically identified as a forged trajectory. By comparing the calculated outlier with the preset threshold, the system can directly output the judgment result, thereby greatly reducing the workload of manual review and meeting the application requirements of real-time monitoring.

[0020] In some technical solutions, the ship information may optionally include one or more combinations of ship type, ship length, and ship width.

[0021] In this technical solution, ship information can include ship type, ship length, and ship width, etc. By introducing key attributes such as ship type and size, this invention enables the graph autoencoder model to effectively distinguish the inherent and differentiated navigation behavior patterns of different types of ships, avoiding misjudging normal behavioral differences between different types of ships as abnormal, and greatly enhancing the scientific rigor and accuracy of collaborative behavior pattern learning.

[0022] According to a second aspect of the present invention, a system for identifying ship spoofed trajectories is proposed, comprising: an acquisition module for acquiring AIS data and a graph autoencoder model; an extraction module for extracting the ship trajectory and ship information of each ship from the AIS data; a first processing module for constructing a cooperative behavior graph among multiple ships based on the ship trajectory and ship information of each ship; a second processing module for determining outliers in each ship trajectory based on the cooperative behavior graph and the graph autoencoder model; and a third processing module for determining spoofed trajectories based on the outliers in each ship trajectory.

[0023] The ship trajectory spoofing identification system provided by this invention mainly includes: an acquisition module, an extraction module, a first processing module, a second processing module, and a third processing module. The acquisition module first acquires AIS (Automatic Identification System) data containing information about multiple ships and a pre-trained graph autoencoder model. The graph autoencoder model is an unsupervised learning model based on graph neural networks, aiming to learn a low-dimensional representation of the graph through the structure of the encoder and decoder, and reconstruct the adjacency matrix of the graph. Subsequently, the extraction module extracts the ship trajectory and ship information for each ship from the mixed AIS data. Extracting the ship trajectory for each ship can be achieved by first determining the overall navigation trajectory of each ship, then segmenting the overall navigation trajectory according to a preset time window to obtain trajectory segments for each ship under different time windows, and finally integrating them according to the time windows. In other words, the ship trajectory mentioned in this invention refers to the ship trajectories of different ships under the same time window. Ship information refers to the static attribute information of the ship, such as ship type, ship name, and ship length. After obtaining the ship trajectory and information of each vessel, the first processing module constructs a collaborative behavior graph among multiple vessels based on their trajectories and information. In other words, it models the behavioral interactions of the vessel group as a graph structure, containing the collaborative patterns of different vessels during navigation. The second processing module then inputs the constructed collaborative behavior graph into a pre-trained graph autoencoder model to obtain outliers corresponding to each vessel's trajectory. The graph autoencoder model can perform unsupervised learning of the collaborative behavior graph, calculating the reconstruction error, i.e., the outlier, of each vessel's trajectory through its encoding-decoding mechanism. Finally, the third processing module automatically identifies abnormal or forged trajectories that deviate from the normal collaborative pattern based on the outliers corresponding to each vessel's trajectory; that is, it determines forged trajectories among multiple vessel trajectories based on the outliers of each vessel's trajectory. In this invention, because the graph autoencoder model learns the universal "collaborative behavior" patterns among vessel groups, rather than local abnormal patterns specific to a particular sea area, it possesses excellent cross-regional and cross-time-time transfer and generalization energy, overcoming the performance degradation problem after scene transitions in related technologies. Meanwhile, in this invention, the detection results are derived from the reconstruction error, i.e., outliers, based on the ship trajectory and the group's cooperative pattern, thus providing a direct explanation for the causes of the anomalies. This invention constructs a cooperative behavior map among ships and utilizes a graph autoencoder model to automatically learn the cooperative patterns of normal ship groups, thereby identifying forged trajectories. This completely eliminates the reliance on difficult-to-obtain labeled data for "forged trajectories," allowing direct detection using massive amounts of unlabeled AIS data. This solves the core pain points of supervised learning methods, such as high labeling costs and a scarcity of forged samples.

[0024] According to a third aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for identifying ship spoofing tracks as described above.

[0025] The electronic device provided by this invention, when the processor executes the computer program, implements the steps of the above-mentioned method for identifying ship spoofing tracks, and can achieve the technical effects of any of the above technical solutions, which will not be elaborated here.

[0026] According to a fourth aspect of the invention, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the steps of the method for identifying ship spoofing tracks as described above.

[0027] The storage medium provided by this invention, when the computer program is executed by the processor, implements the steps of the above-described method for identifying ship spoofing trajectories, and can achieve the technical effects of any of the above technical solutions, which will not be elaborated here.

[0028] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0030] Figure 1 A flowchart illustrating a method for identifying ship spoofing trajectories according to an embodiment of the present invention is shown.

[0031] Figure 2 The diagram illustrates a step in a method for identifying ship spoofing trajectories according to an embodiment of the present invention, which involves constructing a collaborative behavior map among multiple ships based on the ship trajectory and ship information of each ship.

[0032] Figure 3 This diagram illustrates a step in calculating the spatial similarity of the trajectory between any two nodes in a method for identifying ship spoofed trajectories according to an embodiment of the present invention.

[0033] Figure 4 The diagram illustrates a step in a method for identifying ship spoofed trajectories according to an embodiment of the present invention, which involves determining outliers for each ship trajectory based on a cooperative behavior graph and a graph autoencoder model.

[0034] Figure 5This diagram illustrates a step in a method for identifying ship spoofing trajectories according to an embodiment of the present invention, which involves determining the node embedding vector corresponding to a node based on the feature vector of a node, the feature vectors of other nodes connected to the node, the edge feature vector between two nodes, and the encoder.

[0035] Figure 6 The diagram illustrates a step in a method for identifying fake ship trajectories according to an embodiment of the present invention, which involves determining a fake trajectory from multiple ship trajectories based on anomalies in each ship trajectory.

[0036] Figure 7 A structural block diagram of a ship spoofing trajectory identification system according to an embodiment of the present invention is shown. Detailed Implementation

[0037] To better understand the above-mentioned objectives, features, and advantages of the present invention, 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 of the present invention and the features thereof can be combined with each other.

[0038] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0039] Figure 1 A flowchart illustrating a method for identifying ship forged trajectories according to an embodiment of the present invention is shown, wherein the method for identifying ship forged trajectories includes:

[0040] Step 102: Obtain AIS data and graph autoencoder model;

[0041] Step 104: Extract the ship trajectory and ship information for each vessel from the AIS data;

[0042] Step 106: Construct a collaborative behavior map among multiple ships based on the ship trajectory and ship information of each ship;

[0043] Step 108: Determine the outliers for each ship's trajectory based on the cooperative behavior map and the graph autoencoder model;

[0044] Step 110: Determine the fake trajectory based on the outliers of each ship's trajectory.

[0045] The method for identifying spoofed ship trajectories provided by this invention mainly includes: first, acquiring AIS (Automatic Identification System) data containing information about multiple ships and a pre-trained graph autoencoder model. The graph autoencoder model is an unsupervised learning model based on graph neural networks, aiming to learn a low-dimensional representation of the graph through the structure of the encoder and decoder, and reconstruct the adjacency matrix of the graph. Then, extracting the ship trajectory and ship information for each ship from the mixed AIS data. Extracting the ship trajectory for each ship can involve first determining the overall navigation trajectory of each ship, then segmenting the overall navigation trajectory according to a preset time window to obtain trajectory segments for each ship under different time windows, and finally integrating them according to the time windows. In other words, the ship trajectory of each ship mentioned in this invention refers to the ship trajectories of different ships under the same time window. Ship information refers to the static attribute information of the ship, such as ship type, ship name, and ship length. After obtaining the ship trajectory and information of each vessel, a collaborative behavior graph among multiple vessels is constructed based on these parameters. In other words, the behavioral interactions of the vessel group are modeled as a graph structure, containing the collaborative patterns of different vessels during navigation. The constructed collaborative behavior graph is then input into a pre-trained graph autoencoder model to obtain outliers corresponding to each vessel trajectory. The graph autoencoder model can perform unsupervised learning of the collaborative behavior graph, calculating the reconstruction error, i.e., the outlier, of each vessel trajectory through its encoding-decoding mechanism. Finally, based on the outliers corresponding to each vessel trajectory, abnormal or forged trajectories deviating from the normal collaborative pattern are automatically identified; that is, forged trajectories are determined from multiple vessel trajectories based on the outliers of each vessel trajectory. In this invention, because the graph autoencoder model learns the universal "collaborative behavior" patterns among vessel groups, rather than local anomaly patterns specific to a particular sea area, it possesses excellent cross-regional and cross-time-time transfer and generalization energy, overcoming the performance degradation problem after scene transformation in related technologies. Furthermore, in this invention, the detection results are derived from the reconstruction error, i.e., the outlier, of the vessel trajectory and the group's collaborative pattern, thus providing a direct explanation of the causes of anomalies. This invention constructs a cooperative behavior map among ships and uses a graph autoencoder model to automatically learn the cooperative patterns of normal ship groups, thereby identifying forged trajectories. It completely eliminates the dependence on hard-to-obtain labeled data for "forged trajectories" and can directly use massive amounts of unlabeled AIS data for detection, solving the core pain points of high labeling costs and scarcity of forged samples in supervised learning methods.

[0046] Figure 2The diagram illustrates a step in a method for identifying spoofed ship trajectories according to an embodiment of the present invention: constructing a collaborative behavior map among multiple ships based on the ship trajectory and information of each ship. The step of constructing the collaborative behavior map among multiple ships based on the ship trajectory and information of each ship includes:

[0047] Step 202: Use the ship trajectories of multiple ships as multiple nodes;

[0048] Step 204: Count the number of times each ship's trajectory is received at different base stations;

[0049] Step 206: Determine the feature vector of each node based on the number of receptions and ship information;

[0050] Step 208: Calculate the trajectory space similarity between any two nodes;

[0051] Step 210: Determine multiple edges based on the trajectory space similarity between any two nodes;

[0052] Step 212: Calculate the hull difference between the two nodes connected to the edge based on the ship information of each ship;

[0053] Step 214: Use the trajectory space similarity and hull difference between the two nodes connected to the edge as the edge feature vector;

[0054] Step 216: Construct a collaborative behavior graph based on nodes, edges, node feature vectors, and edge feature vectors.

[0055] In this embodiment, the step of constructing a collaborative behavior graph among multiple ships based on the ship trajectory and ship information of each ship includes: first, using the ship trajectory of each ship as a node in the collaborative behavior graph; then, counting the number of times each ship trajectory is received at different base stations, and using the number of receptions as a reception vector reflecting the ship's communication mode; then, determining the feature vector of each node based on the number of receptions and ship information, that is, combining the reception vector with the ship's static attribute information as the node's feature vector; simultaneously, calculating the trajectory spatial similarity between any two nodes, and determining an edge based on the trajectory spatial similarity between any two nodes. For example, when the trajectory spatial similarity between any two nodes is greater than a preset similarity, an edge is established between these two nodes, connecting them; if the trajectory spatial similarity between any two nodes is not greater than the preset similarity, no edge is established between them. By establishing an edge between two nodes with a trajectory spatial similarity greater than the preset similarity, the collaborative relationship between ships is represented. Then, based on the ship information of each ship, the ship difference between the two nodes connected by the edge is calculated, and then the trajectory spatial similarity and the ship hull difference are used as the feature vector of this edge, i.e., the edge feature vector. In this invention, each edge is assigned a feature vector. This feature vector is composed of the trajectory spatial similarity between the two nodes and the corresponding ship hull differences, thus precisely quantifying the strength and reliability of the cooperative relationship. Ultimately, through these four elements—nodes, edges, node feature vectors, and edge feature vectors—a complete graph capable of deeply expressing the multi-dimensional cooperative behavior of ship groups across space, signals, and attributes is constructed. By introducing the receiving vector and ship hull differences, this invention expands the cooperative relationship from a single spatial similarity to multiple dimensions such as communication behavior matching degree and ship physical attribute compatibility, resulting in a richer information content in the constructed graph and a more accurate description of cooperative behavior.

[0056] Figure 3 The diagram illustrates a step in calculating the trajectory spatial similarity between any two nodes in a method for identifying ship spoofed trajectories according to an embodiment of the present invention. The step of calculating the trajectory spatial similarity between any two nodes includes:

[0057] Step 302: Calculate the Hausdorff distance between any two nodes based on the spatial position of each ship's trajectory;

[0058] Step 304: Determine the trajectory space similarity based on Hausdorff distance.

[0059] In this embodiment, the step of calculating the trajectory spatial similarity between any two nodes includes: firstly, calculating the Hausdorff distance between any two nodes based on the spatial position of each ship trajectory; and then determining the trajectory spatial similarity based on the Hausdorff distance. In other words, this invention uses the Hausdorff distance as the core algorithm for calculating the trajectory spatial similarity between any two ship trajectories. By calculating the maximum and minimum distances between all points on one ship trajectory and all points on another ship trajectory, it accurately measures the overall matching degree and consistency of the two ship trajectories in the entire spatial morphology, thereby transforming the abstract "spatial similarity" into a computable and stable data metric.

[0060] Figure 4 The diagram illustrates a flowchart of a method for identifying forged ship trajectories according to an embodiment of the present invention, showing the step of determining outliers for each ship trajectory based on a cooperative behavior graph and a graph autoencoder model. The graph autoencoder model includes an encoder and a decoder. The step of determining outliers for each ship trajectory based on the cooperative behavior graph and the graph autoencoder model includes:

[0061] Step 402: Determine the original adjacency matrix for each node based on the cooperative behavior graph;

[0062] Step 404: Determine the node embedding vector corresponding to the node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder;

[0063] Step 406: Determine the reconstructed adjacency matrix for each node based on the node embedding vector and the decoder;

[0064] Step 408: Determine the outliers for each node based on the original adjacency matrix and the reconstructed adjacency matrix.

[0065] In this embodiment, the graph autoencoder model includes an encoder and a decoder. The steps for determining outliers in each ship trajectory based on the cooperative behavior graph and the graph autoencoder model include: First, extracting the original adjacency matrix representing the true connections between nodes from the cooperative behavior graph; then, inputting the information from the cooperative behavior graph—namely, the feature vectors of nodes, the feature vectors of other nodes connected to a node, and the edge feature vectors between two nodes—into the encoder. The encoder maps each node to a low-dimensional node embedding vector containing contextual information about itself and its neighbors (i.e., other nodes connected to the node) through an information aggregation mechanism. Next, the decoder reconstructs the graph's connection relationships using these node embedding vectors through operations such as inner product, thus generating a reconstructed adjacency matrix. Finally, the difference between the corresponding row (or column) vectors of each node in the original adjacency matrix and the reconstructed adjacency matrix is ​​calculated, and this difference is used as the outlier of that node. This invention, through an "encoding-decoding-comparison" process, transforms the learning of "normal" cooperative behavior patterns into the ability to reconstruct graph structures, automatically identifying nodes that cannot be accurately reconstructed by the model as outliers, thereby achieving the detection of forged trajectories without any manual annotation.

[0066] Figure 5 The diagram illustrates a step in a method for identifying ship spoofed trajectories according to an embodiment of the present invention: determining the node embedding vector corresponding to a node based on the feature vector of a node, the feature vectors of other nodes connected to the node, the edge feature vector between two nodes, and the encoder. The step of determining the node embedding vector corresponding to a node based on the feature vector of a node, the feature vectors of other nodes connected to the node, the edge feature vector between two nodes, and the encoder includes:

[0067] Step 502: Determine the attention weights for each other node based on the node's feature vector, the feature vectors of other nodes connected to the node, and the edge feature vectors between the two nodes;

[0068] Step 504: Determine the node embedding vector corresponding to each node based on the attention weights of each other node, the feature vector of the node, and the encoder.

[0069] In this embodiment, the step of determining the node embedding vector corresponding to a node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vectors between two nodes, and the encoder includes: First, for the current central node, its own feature vector, the feature vectors of each of its neighboring nodes, and the edge feature vectors connecting them are comprehensively considered to dynamically calculate the attention weight for each neighboring node. Here, a neighboring node refers to a node connected to the central node via an edge, and the central node is any node in the cooperative behavior graph. The attention weight can accurately quantify the importance of different neighboring nodes to the central node. Subsequently, the encoder does not aggregate all neighboring node information equally, but rather performs a weighted summation of the feature vectors of each neighboring node based on the calculated attention weight. The weighted summation feature vector is then fused with the central node's feature vector to form a node embedding vector that more accurately reflects the central node's behavior in its specific cooperative context. By determining the attention weight corresponding to each neighboring node, this invention prevents the model from treating all spatially similar neighboring nodes equally, instead focusing on those neighboring nodes with highly correlated behaviors and a high probability of cooperative interaction, thereby significantly improving the quality and relevance of information aggregation.

[0070] Figure 6 The diagram illustrates a step in a method for identifying forged ship trajectories according to an embodiment of the present invention: determining the forged trajectory from multiple ship trajectories based on outliers of each ship trajectory. The step of determining the forged trajectory from multiple ship trajectories based on outliers of each ship trajectory includes:

[0071] Step 602: Compare the outliers of each ship trajectory with a preset threshold;

[0072] Step 604: When the abnormal value of the ship trajectory is greater than the preset threshold, the ship trajectory is determined to be a fake trajectory.

[0073] In this embodiment, the step of identifying a forged trajectory from multiple ship trajectories based on the outlier of each ship trajectory includes: firstly, comparing the outlier of each ship trajectory with a preset threshold; and secondly, automatically identifying a ship trajectory as a forged trajectory when the outlier of a particular ship trajectory exceeds the preset threshold. By comparing the calculated outlier with the preset threshold, the system can directly output the determination result, thereby greatly reducing the workload of manual review and meeting the application requirements of real-time monitoring.

[0074] In some embodiments, the ship information may optionally include one or more combinations of ship type, ship length, and ship width.

[0075] In this embodiment, ship information can include ship type, ship length, and ship width, etc. By introducing key attributes such as ship type and size, this invention enables the graph autoencoder model to effectively distinguish the inherent and differentiated navigation behavior patterns of different types of ships, avoiding misjudging normal behavioral differences between different types of ships as abnormal, and greatly enhancing the scientific rigor and accuracy of collaborative behavior pattern learning.

[0076] For example, the ship spoofing trajectory identification system provided by this invention mainly includes: a data preparation module, a cooperative graph construction module, a graph autoencoder module, and an anomaly detection module. The data preparation module is mainly used to extract ship trajectories from raw AIS data, then divide the trajectory segments according to time windows, and simultaneously calculate the received vector (an 800-dimensional vector, because it involves 800 base stations, each dimension being the signal reception frequency of the base station) and ship features (ship length, ship width, and ship type). The cooperative graph construction module calculates the similarity between trajectories based on Hausdorff distance, establishes a cooperative behavior graph, and generates node features and edge features. The graph autoencoder module mainly includes a GAT (Graph Attention Encoder) encoder and an adjacency reconstruction decoder, learning the latent structure of cooperative relationships between nodes. The anomaly detection module can calculate node reconstruction errors and output an anomaly score. The process of this invention is as follows: first, a time sliding window is constructed, generating a 3-hour time sliding window every hour; then, for each ship, the trajectory point sequence and signal reception count within this time window are extracted, and an input vector is formed based on the signal reception count and ship features.

[0077] x i =[r i,1 ,r i,2 ,……,r i,m ,type i ,length i ,width i ];

[0078] Where i represents the ship, m represents the base station, and r i,m For the number of times ship i receives data at base station m, type i For ship i, the ship type is length. i Let width be the length of ship i. i Let x be the beam of ship i. i Let i represent the input vector of ship i.

[0079] A cooperative behavior graph is then generated. In the cooperative behavior graph, nodes represent time windows for each ship. The feature vectors of the edges are then calculated using the following formula:

[0080] wij =exp(-d Haus (i, j) / α);

[0081] Where i and j represent nodes, d Haus Let be the Haudorff distance between any two ships within a time window, and exp() represent an exponential function with base e. α is a decay coefficient used to control the degree of influence of distance on edge weights, and w ij The feature vector represents the edge between node i and node j.

[0082] The process then utilizes a GAT encoder. For example, firstly, node features are mapped to a unified latent space; then, attention energy is calculated for each edge, normalized to obtain attention weights; finally, a new node vector z is obtained based on these multiple attention weights. i .

[0083] Then, self-encoding reconstruction is performed. First, the adjacency matrix is ​​reconstructed using the decoder through node embedding, as shown in the formula: ;

[0084] Among them, z i and z j Let σ represent a new vector between two adjacent nodes, and let σ() be the Sigmoid (activation) function. Let T represent the reconstructed adjacency matrix, and let T represent the permutation matrix.

[0085] The training objective is then to minimize the reconstruction error of the adjacency matrix:

[0086] ;

[0087] in, Indicates error, A i,j Represents the original adjacency matrix. This indicates the reconstruction of the adjacency matrix, and Σ represents the summation symbol, which means accumulating all items within a given index range.

[0088] Finally, anomaly scores are calculated. For example, an anomaly score is defined for each node i:

[0089] ;

[0090] Among them, score i A represents an outlier score. i Represents the original adjacency matrix. This indicates the reconstruction of the adjacency matrix. This represents the L2 norm, i.e., the Euclidean distance.

[0091] A high score indicates that the cooperative behavior of this node cannot be explained by the model, so the ship trajectory corresponding to this node can be regarded as a fake trajectory.

[0092] Figure 7 A schematic block diagram of a ship spoofing trajectory identification system according to an embodiment of the present invention is shown, wherein the ship spoofing trajectory identification system 70 includes:

[0093] Module 702 is used to acquire AIS data and graph autoencoder model;

[0094] Extraction module 704 is used to extract the ship trajectory and ship information of each ship from AIS data;

[0095] The first processing module 706 is used to construct a collaborative behavior map among multiple ships based on the ship trajectory and ship information of each ship.

[0096] The second processing module 708 is used to determine the outliers of each ship trajectory based on the cooperative behavior map and the graph autoencoder model;

[0097] The third processing module 710 is used to determine the fake trajectory based on the abnormal values ​​of each ship trajectory.

[0098] The ship spoofing trajectory identification system 70 provided by this invention mainly includes: an acquisition module 702, an extraction module 704, a first processing module 706, a second processing module 708, and a third processing module 710. The acquisition module 702 first acquires AIS (Automatic Identification System) data containing information about multiple ships and a pre-trained graph autoencoder model. The graph autoencoder model is an unsupervised learning model based on graph neural networks, aiming to learn a low-dimensional representation of the graph through the structure of the encoder and decoder, and reconstruct the adjacency matrix of the graph. Subsequently, the extraction module 704 extracts the ship trajectory and ship information of each ship from the mixed AIS data. Extracting the ship trajectory of each ship can be achieved by first determining the overall navigation trajectory of each ship, then segmenting the overall navigation trajectory according to a preset time window to obtain trajectory segments of each ship under different time windows, and finally integrating them according to the time windows. In other words, the ship trajectory of each ship mentioned in this invention refers to the ship trajectories of different ships under the same time window. Ship information refers to the static attribute information of the ship, such as ship type, ship name, and ship length. After obtaining the ship trajectory and information of each vessel, the first processing module 706 constructs a collaborative behavior graph among multiple vessels based on their trajectories and information. In other words, it models the behavioral interactions of the vessel group as a graph structure, containing the collaborative patterns of different vessels during navigation. Subsequently, the second processing module 708 inputs the constructed collaborative behavior graph into a pre-trained graph autoencoder model to obtain outliers corresponding to each vessel trajectory. The graph autoencoder model can perform unsupervised learning of the collaborative behavior graph, calculating the reconstruction error, i.e., the outlier, of each vessel trajectory through its encoding-decoding mechanism. Finally, the third processing module 710 automatically identifies abnormal or forged trajectories that deviate from the normal collaborative pattern based on the outliers corresponding to each vessel trajectory; that is, it determines forged trajectories among multiple vessel trajectories based on the outliers of each vessel trajectory. In this invention, because the graph autoencoder model learns the universal "collaborative behavior" patterns among vessel groups, rather than local abnormal patterns in a specific sea area, it possesses excellent cross-regional and cross-time-time migration and generalization energy, overcoming the performance degradation problem after scene transitions in related technologies. Meanwhile, in this invention, the detection results are derived from the reconstruction error, i.e., outliers, based on the ship trajectory and the group's cooperative pattern, thus providing a direct explanation for the causes of the anomalies. This invention constructs a cooperative behavior map among ships and utilizes a graph autoencoder model to automatically learn the cooperative patterns of normal ship groups, thereby identifying forged trajectories. This completely eliminates the reliance on difficult-to-obtain labeled data for "forged trajectories," allowing direct detection using massive amounts of unlabeled AIS data. This solves the core pain points of supervised learning methods, such as high labeling costs and a scarcity of forged samples.

[0099] In some embodiments, optionally, the first processing module 706 is specifically configured to: treat the ship trajectories of multiple ships as multiple nodes; count the number of times each ship trajectory is received at different base stations; determine the feature vector of each node based on the number of receptions and ship information; calculate the trajectory spatial similarity between any two nodes; determine multiple edges based on the trajectory spatial similarity between any two nodes; calculate the hull difference between two nodes connected to the edge based on the ship information of each ship; use the trajectory spatial similarity and hull difference between two nodes connected to the edge as edge feature vectors; and construct a cooperative behavior graph based on the nodes, edges, node feature vectors, and edge feature vectors.

[0100] In this embodiment, the first processing module 706 is specifically used to first treat the ship trajectory of each vessel as a node in the cooperative behavior graph, then count the number of times each vessel trajectory is received at different base stations, and use the number of receptions as a reception vector reflecting the ship's communication mode. Then, based on the number of receptions and ship information, the feature vector of each node is determined, that is, the reception vector is combined with the ship's static attribute information as the node's feature vector. Simultaneously, the trajectory spatial similarity between any two nodes is calculated, and an edge is determined based on the trajectory spatial similarity between any two nodes. For example, when the trajectory spatial similarity between any two nodes is greater than a preset similarity, an edge is established between these two nodes, connecting them; if the trajectory spatial similarity between any two nodes is not greater than the preset similarity, no edge is established between them. By establishing an edge between two nodes with a trajectory spatial similarity greater than the preset similarity, the cooperative relationship between ships is represented. Then, based on the ship information of each vessel, the ship difference between the two nodes connected by the edge is calculated, and then the trajectory spatial similarity and the ship hull difference are used as the feature vector of this edge, i.e., the edge feature vector. In this invention, each edge is assigned a feature vector. This feature vector is composed of the trajectory spatial similarity between the two nodes and the corresponding ship hull differences, thus precisely quantifying the strength and reliability of the cooperative relationship. Ultimately, through these four elements—nodes, edges, node feature vectors, and edge feature vectors—a complete graph capable of deeply expressing the multi-dimensional cooperative behavior of ship groups across space, signals, and attributes is constructed. By introducing the receiving vector and ship hull differences, this invention expands the cooperative relationship from a single spatial similarity to multiple dimensions such as communication behavior matching degree and ship physical attribute compatibility, resulting in a richer information content in the constructed graph and a more accurate description of cooperative behavior.

[0101] In some embodiments, optionally, the first processing module 706 is further configured to calculate the Hausdorff distance between any two nodes based on the spatial location of each ship trajectory; and determine the trajectory spatial similarity based on the Hausdorff distance.

[0102] In this embodiment, the first processing module 706 is further specifically used to first calculate the Hausdorff distance between any two nodes based on the spatial position of each ship trajectory, and then determine the trajectory spatial similarity based on the Hausdorff distance. That is, in this invention, the Hausdorff distance is used as the core algorithm for calculating the trajectory spatial similarity between any two ship trajectories. By calculating the maximum and minimum distances between all points on one ship trajectory and all points on another ship trajectory, the overall matching degree and consistency of the two ship trajectories in the entire spatial morphology is accurately measured, thereby transforming the abstract "spatial similarity" into a computable and stable data metric.

[0103] In some embodiments, optionally, the graph autoencoder model includes an encoder and a decoder, and the second processing module 708 is specifically configured to: determine the original adjacency matrix corresponding to each node based on the cooperative behavior graph; determine the node embedding vector corresponding to the node based on the node's feature vector, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder; determine the reconstructed adjacency matrix corresponding to each node based on the node embedding vector and the decoder; and determine the outliers of each node based on the original adjacency matrix and the reconstructed adjacency matrix.

[0104] In this embodiment, the graph autoencoder model includes an encoder and a decoder. The second processing module 708 is specifically used to extract the original adjacency matrix representing the true connection relationships between nodes from the cooperative behavior graph. Then, it inputs the information from the cooperative behavior graph—namely, the feature vectors of the nodes, the feature vectors of other nodes connected to the node, and the edge feature vectors between two nodes—into the encoder. The encoder maps each node to a low-dimensional node embedding vector containing contextual information about itself and its neighbors (i.e., other nodes connected to the node) through an information aggregation mechanism. Next, the decoder reconstructs the connection relationships of the graph using these node embedding vectors through operations such as inner product, thus generating a reconstructed adjacency matrix. Finally, by calculating the difference between the corresponding row (or column) vectors of each node in the original adjacency matrix and the reconstructed adjacency matrix, this difference is used as the outlier value of that node. This invention, through an "encoding-decoding-comparison" process, transforms the learning of "normal" cooperative behavior patterns into the ability to reconstruct graph structures, automatically identifying nodes that cannot be accurately reconstructed by the model as outliers, thereby achieving the discovery of forged trajectories without any manual annotation.

[0105] In some embodiments, the second processing module 708 is optionally further configured to determine the attention weight corresponding to each other node based on the feature vector of the node, the feature vectors of other nodes connected to the node, and the edge feature vector between the two nodes; and to determine the node embedding vector corresponding to the node based on the attention weight of each other node, the feature vector of the node, and the encoder.

[0106] In this embodiment, the second processing module 708 is further specifically used to dynamically calculate the attention weight for each neighbor node by comprehensively considering its own feature vector, the feature vectors of each of its neighbor nodes, and the feature vectors of the edges connecting them. Here, a neighbor node refers to a node connected to the center node via an edge, and the center node is any node in the cooperative behavior graph. The attention weight can accurately quantify the importance of different neighbor nodes to the center node. Subsequently, the encoder does not aggregate all neighbor node information equally, but rather performs a weighted summation of the feature vectors of each neighbor node based on the calculated attention weight. The weighted summation feature vector is then fused with the feature vector of the center node to form a node embedding vector that more accurately reflects the behavior of the center node in its specific cooperative context. By determining the attention weight corresponding to each neighbor node, this invention prevents the model from treating all spatially similar neighbor nodes equally, instead focusing on those neighbor nodes with highly correlated behaviors and a high probability of cooperative interaction, thereby significantly improving the quality and relevance of information aggregation.

[0107] In some embodiments, the third processing module 710 is optionally configured to compare the outlier value of each ship trajectory with a preset threshold; and determine that the ship trajectory is a fake trajectory when the outlier value of the ship trajectory is greater than the preset threshold.

[0108] In this embodiment, the third processing module 710 is specifically used to first compare the abnormal value of each ship trajectory with a preset threshold. When the abnormal value of a ship trajectory exceeds the preset threshold, the ship trajectory is automatically identified as a forged trajectory. By comparing the calculated abnormal value with the preset threshold, the system can directly output the judgment result, thereby greatly reducing the workload of manual review and meeting the application requirements of real-time monitoring.

[0109] In some embodiments, the ship information may optionally include one or more combinations of ship type, ship length, and ship width.

[0110] In this embodiment, ship information can include ship type, ship length, and ship width, etc. By introducing key attributes such as ship type and size, this invention enables the graph autoencoder model to effectively distinguish the inherent and differentiated navigation behavior patterns of different types of ships, avoiding misjudging normal behavioral differences between different types of ships as abnormal, and greatly enhancing the scientific rigor and accuracy of collaborative behavior pattern learning.

[0111] An electronic device according to an embodiment of the present invention includes a memory processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying ship spoofing tracks as described above.

[0112] The electronic device provided by the present invention, when the processor executes the computer program, implements the steps of the above-described method for identifying ship spoofing trajectories, and can achieve the technical effects of any of the above embodiments, which will not be repeated here.

[0113] One embodiment of the present invention provides a storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method for identifying ship spoofing tracks as described above.

[0114] The storage medium provided by this invention, when the computer program is executed by a processor, implements the steps of the above-described method for identifying ship spoofing trajectories, and can achieve the technical effects of any of the above embodiments, which will not be repeated here.

[0115] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance, unless otherwise expressly specified and limited. The terms "connection," "installation," and "fixing," etc., should be interpreted broadly. For example, "connection" can mean a fixed connection, a detachable connection, or an integral connection; it can mean 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.

[0116] 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.

[0117] 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 identifying forged ship tracks, characterized in that, include: Acquire AIS data and graph autoencoder model; The ship trajectory and ship information of each ship are extracted from the AIS data, wherein the ship information includes one or more combinations of ship type, ship length and ship width; A collaborative behavior map among multiple vessels is constructed based on the vessel trajectory and vessel information of each vessel. Anomalies in each of the ship trajectories are determined based on the cooperative behavior map and the graph autoencoder model; The forged trajectory is determined based on the outlier of each of the ship trajectories; The step of constructing a cooperative behavior map among multiple vessels based on the vessel trajectory and vessel information of each vessel includes: The ship trajectories of the multiple ships are used as multiple nodes; The number of times each ship trajectory was received at different base stations was counted. The feature vector of each node is determined based on the number of receptions and the ship information; Calculate the trajectory space similarity between any two of the nodes; Multiple edges are determined based on the trajectory space similarity between any two nodes. When the trajectory space similarity between any two nodes is greater than a preset similarity, an edge is established between the two nodes to represent the cooperative relationship between the ships. The hull difference between the two nodes connected to the edge is calculated based on the ship information of each ship. The trajectory space similarity and the hull difference between the two nodes connected to the edge are used as edge feature vectors; The cooperative behavior graph is constructed based on the nodes, the edges, the feature vectors of the nodes, and the feature vectors of the edges. The graph autoencoder model includes an encoder and a decoder. The step of determining outliers for each ship trajectory based on the cooperative behavior graph and the graph autoencoder model includes: Based on the cooperative behavior graph, determine the original adjacency matrix representing the actual connection relationship between the nodes for each node; The node embedding vector corresponding to the node is determined based on the feature vector of the node, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder. The reconstructed adjacency matrix corresponding to each node is determined based on the node embedding vector and the decoder; Calculate the difference between the corresponding row or column vectors of each node in the original adjacency matrix and the reconstructed adjacency matrix, and use this difference as the outlier of the node.

2. The method for identifying ship spoofing tracks according to claim 1, characterized in that, The step of calculating the trajectory space similarity between any two nodes includes: The Hausdorff distance between any two nodes is calculated based on the spatial position of each of the ship trajectories. The trajectory space similarity is determined based on the Hausdorff distance.

3. The method for identifying ship spoofing tracks according to claim 1, characterized in that, The step of determining the node embedding vector corresponding to the node based on the feature vector of the node, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder includes: The attention weights for each of the other nodes are determined based on the feature vectors of the node, the feature vectors of the other nodes connected to the node, and the feature vectors of the edges between the two nodes. The node embedding vector corresponding to the node is determined based on the attention weight of each of the other nodes, the feature vector of the node, and the encoder.

4. The method for identifying ship spoofing tracks according to any one of claims 1 to 3, characterized in that, The step of determining the forged trajectory based on the outlier of each of the ship trajectories includes: The outlier value of each ship trajectory is compared with a preset threshold; When the abnormal value of the ship trajectory is greater than the preset threshold, the ship trajectory is determined to be the forged trajectory.

5. A system for identifying spoofed ship tracks, characterized in that, include: The acquisition module is used to acquire AIS data and a graph autoencoder model; The extraction module is used to extract the ship trajectory and ship information of each ship from the AIS data, wherein the ship information includes one or more combinations of ship type, ship length and ship width; A first processing module is configured to construct a collaborative behavior map among multiple vessels based on the vessel trajectory and vessel information of each vessel. The second processing module is used to determine the outlier values ​​of each ship trajectory based on the cooperative behavior map and the graph autoencoder model. The third processing module is used to determine the forged trajectory based on the abnormal value of each ship trajectory; The first processing module is specifically used to treat the ship trajectories of the multiple ships as multiple nodes; count the number of times each ship trajectory is received at different base stations; determine the feature vector of each node based on the number of receptions and the ship information; calculate the trajectory spatial similarity between any two nodes; determine multiple edges based on the trajectory spatial similarity between any two nodes, and when the trajectory spatial similarity between any two nodes is greater than a preset similarity, establish an edge between the two nodes to represent the cooperative relationship between the ships; calculate the hull difference between the two nodes connected to the edge based on the ship information of each ship; use the trajectory spatial similarity and hull difference between the two nodes connected to the edge as edge feature vectors; and construct the cooperative behavior graph based on the nodes, the edges, the feature vectors of the nodes, and the edge feature vectors. The graph autoencoder model includes an encoder and a decoder. The second processing module is specifically used to determine the original adjacency matrix representing the real connection relationship between the nodes for each node based on the cooperative behavior graph; determine the node embedding vector corresponding to the node based on the feature vector of the node, the feature vectors of other nodes connected to the node, the edge feature vector between the two nodes, and the encoder; determine the reconstructed adjacency matrix corresponding to each node based on the node embedding vector and the decoder; calculate the difference between the corresponding row or column vectors of each node in the original adjacency matrix and the reconstructed adjacency matrix, and use this difference as the outlier of the node.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for identifying ship spoofing tracks as described in any one of claims 1 to 4.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for identifying ship spoofing tracks as described in any one of claims 1 to 4.

Images