Ship automatic prevention and control tracking monitoring system based on AIS, radar and video signal fusion

Abstract

The present application relates to the technical fields of ship tracking and monitoring, and discloses an automatic prevention and control tracking and monitoring system for ships based on AIS, radar and video signal fusion, comprising: a video signal capturing unit; a signal fusion trajectory prediction unit, which interacts with the video signal capturing unit, is used for fusing ship digital signals, radar digital signals and video signals, and predicting the ship running trajectory in the target period based on the signal fusion result; a track compliance evaluation unit, which is used for obtaining the ship's to-be-traveled route, calculating the offset distance between the ship running trajectory and the ship's to-be-traveled route, and evaluating the track compliance degree of the ship in the driving process based on the offset distance; an arrival period prediction unit; and a tracking and monitoring control unit. The present application realizes multi-dimensional and high-precision ship prevention and control tracking and monitoring, and provides comprehensive technical support and guarantee for ship navigation safety, traffic management and operation efficiency improvement.

Description

Technical Field

[0001] This invention relates to the field of ship tracking and monitoring technology, and more specifically, to an automatic ship control and tracking monitoring system based on the fusion of AIS, radar, and video signals. Background Technology

[0002] Automatic ship tracking and monitoring employs technologies such as multi-source sensing and data fusion to achieve automated control of the entire process of ship positioning, trajectory tracking, risk warning, equipment monitoring and emergency response, meeting the needs of maritime compliance and safe operation. Existing tracking and monitoring technologies mainly focus on sensing, processing, decision-making and execution as the core links, forming a complete system covering shipborne terminals, shore-based platforms and application services.

[0003] For example, Chinese patent application number CN104809917B discloses a real-time ship tracking and monitoring method, including: collecting AIS messages and related information sent by the tracked ship, and storing the collected information in the AIS database of the ship tracking system; randomly selecting a ship for monitoring, and using the latest recorded position and speed information of the ship in the database as the initial monitoring value; based on the initial monitoring value, performing coordinate position transformation, speed transformation, and focal length conversion on the AIS information of the target ship to be tracked to obtain relevant transformation values; based on the transformation values, predicting the ship's navigation coordinates and speed at the next moment, verifying the predicted values ​​against control thresholds to determine the tracking PTZ camera; and correcting the control parameters of the tracking PTZ camera.

[0004] However, the aforementioned patents only rely on AIS messages to obtain information such as ship coordinates and speed, without integrating effective data from radar signals and video signals. This results in a single data dimension, making it impossible to form multi-source data cross-verification, limiting monitoring accuracy, and failing to form an efficient and interconnected information loop. Furthermore, it does not consider the environmental conditions of the ship's navigation area, only predicting and monitoring the ship's own position and speed, leading to a lack of scientific planning and timely guidance for ship navigation.

[0005] Furthermore, existing monitoring technologies heavily rely on manual monitoring operations, which are not only susceptible to interference from factors such as the subjective judgment and energy level of monitoring personnel, but also generally suffer from problems such as information transmission delays and high target identification deviation rates. For example, when manually screening abnormal targets, it often takes tens of minutes to complete information verification, making it difficult to meet the real-time prevention and control needs around ships.

[0006] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0007] To address the problems in related technologies, this invention proposes an automatic ship control and tracking monitoring system based on the fusion of AIS, radar, and video signals, in order to overcome the aforementioned technical problems existing in the current related technologies.

[0008] Therefore, the specific technical solution adopted by the present invention is as follows: An automatic ship control and tracking monitoring system based on the fusion of AIS, radar, and video signals, comprising: The video signal capture unit is used to adjust the tracking state of the camera capture device according to the latitude, longitude and angle adjustment technology of the camera capture device, and to capture the video signal of the ship during the navigation process using the camera capture device after the tracking state adjustment. The signal fusion trajectory prediction unit interacts with the video signal capture unit to fuse ship digital signals, radar digital signals and video signals, and predict the ship's trajectory within the target time period based on the signal fusion results. The track compliance assessment unit is used to obtain the route to be navigated by the ship and calculate the offset distance between the ship's operating trajectory and the route to be navigated. Based on the offset distance, the track compliance of the ship during the navigation process is assessed. The arrival time prediction unit is used to adjust the ship's speed according to the compliance of the navigation track, and to assess the expected time when the ship will arrive at the water service area based on the adjusted ship speed and the environmental conditions in the navigation area. The tracking and monitoring control unit is used to acquire the temperature and navigation conditions in the navigation area during the expected period, analyze the consumption rate of the ship's hydraulic oil, and verify and optimize the ship's navigation speed based on the consumption rate to ensure the normal supply of the ship's hydraulic oil. The signal fusion trajectory prediction unit, the trajectory compliance assessment unit, the arrival time prediction unit, and the tracking and monitoring control unit constitute a closed-loop monitoring logic to complete the automatic prevention, control, tracking, and monitoring of ships during navigation.

[0009] Preferably, the signal fusion trajectory prediction unit includes: The digital signal acquisition module is used to acquire the ship's digital signals and radar digital signals during navigation using the Automatic Identification System (AIS) and radar detector, respectively. The signal fusion output module is used to calculate the distance between the ship's digital signal, radar digital signal and video signal, determine the corresponding trust function, and perform signal fusion processing by combining fuzzy clustering technology to output fused signal data. The navigation trajectory prediction module is used to combine fused signal data with ship route maps to output a set of candidate trajectory endpoints, and generate a heat map based on the set of candidate trajectory endpoints to predict the ship's operating trajectory within the target time period.

[0010] Preferably, the distances between the ship's digital signal, radar digital signal, and video signal are calculated, and the corresponding trust functions are determined. Then, fuzzy clustering technology is used to perform signal fusion processing, and the output fused signal data includes: Outlier removal and normalization are performed on ship digital signals, radar digital signals and video signals, and the distance between any two types of signals is calculated after processing to form a distance matrix; The element similarity between ship digital signals, radar digital signals and video signals is calculated based on the distance matrix. After normalizing the element similarity, the trust function between any two types of signals is calculated. Remove signal data from ship digital signals, radar digital signals, and video signals whose trust function is lower than the trust threshold, and calculate the mean of the remaining signal data to output outlier data; Principal component analysis is used to process heterogeneous data, and a similarity matrix between heterogeneous data is calculated based on the processing results, so as to capture the relationship between heterogeneous data using the similarity matrix; Based on the similarity matrix, the clustering radius is set, and the optimal solution is iteratively screened to determine the optimal solution. The optimal solution is then used to calculate the weight vector to generate the membership matrix, which is used to fuse heterogeneous data and output fused signal data.

[0011] Preferably, based on fused signal data and combined with a ship route map, a set of candidate trajectory endpoints is output, and a heat map is generated based on the candidate trajectory endpoint set to predict the ship's operating trajectory within the target time period, including: Based on fused signal data and combined with ship route map information, graph neural network technology is used to jointly encode the ship's trajectory information and ship route map information to obtain scene coding vector; The set of candidate trajectory points of the ship in the navigation area is obtained based on the scene encoding vector, and the probability value of each candidate trajectory point in the set within the target time period is analyzed by combining the attention mechanism. A heat map is generated based on the probability values ​​of each candidate trajectory point, presenting the probability of a ship appearing in the navigation area within the target time period. Based on the occurrence probability and the trajectory completion processing method, the ship's operating trajectory within the target time period is predicted and the ship's operating trajectory is determined.

[0012] Preferably, based on fused signal data and combined with ship route map information, graph neural network technology is used to jointly encode the ship's trajectory information and ship route map information to obtain a scene encoding vector including: The fused signal data is split into trajectory vectors connected end to end in chronological order, and the ship route map information is defined as map vectors to describe the geographical features of ship navigation. By using trajectory vectors and map vectors as graph nodes and adjacent vectors as edges, node information is transferred and local features are extracted to obtain the encoded features of the local ship trajectory map and the local map map. The encoded features are used as input to perform global information fusion through a graph neural network to construct a global scene graph containing the ship's historical trajectory and navigation environment, resulting in a two-dimensional scene encoded feature matrix. The fused signal data is embedded into standardized coordinates to obtain an embedding vector. The embedding vector is used to represent the coordinate encoding information of the ship at different times. The embedding vector is then input into the encoder sequentially according to each time to extract the temporal features of the ship's motion state. The temporal features output by the encoder are input into a fully connected layer containing an activation function to obtain the encoded vector of the ship's motion state.

[0013] Preferably, a set of candidate trajectory points for the ship within the navigation area is obtained based on the scene encoding vector, and the probability value of each candidate trajectory point in the candidate trajectory point set within the target time period is analyzed using an attention mechanism, including: Based on the ship route map information, the navigation area is determined. Within the navigation area, the target prediction time period and the ship's navigation speed operation scenario coding map are combined to determine the ship's trajectory points within the navigation area. Candidate trajectory points are integrated based on a dense sampling method, and candidate points that are more than a threshold distance and conflict with obstacles in the navigation area are removed to obtain a set of candidate trajectory points. The two-dimensional scene coding feature matrix and coding vector are fused. The coding vectors related to ships in the two-dimensional scene coding feature matrix are superimposed, and the row vectors that are not related are filled with zero vectors to obtain the fused matrix. Based on the fusion matrix and the two-dimensional scene coding feature matrix, a comprehensive coding matrix containing information on ship navigation scene and operation status is determined. The comprehensive coding matrix and the candidate trajectory point set are then mapped to a query matrix, a key matrix, and a value matrix using a projection matrix. Based on the query matrix, key matrix, and value matrix, the attention distribution of the association weights between candidate trajectory points and encoding vectors is calculated using the scaling point attention mechanism. Based on the attention distribution, the trajectory driving candidate points after fusing attention features are obtained and input into the fully connected layer to output the probability value of the trajectory driving candidate points in the target time period.

[0014] Preferably, the flight track compliance assessment unit includes: The route division module is used to obtain the route to be navigated by the ship, divide the route to be navigated by the ship into three stages according to the route status, and determine the anchoring feature points of each segment of the route to be navigated by the ship. The three stages include the mandatory segment, the adjustment segment, and the warning segment. The offset distance definition module is used to synchronize and align the ship's trajectory with the segmented results of the ship's route according to time nodes, and define the offset distance between the ship's trajectory and the route to be navigated based on anchor feature points and segment comparison logic. The navigation compliance judgment module is used to assign weights to the offset distance of each segment of the ship's route according to the navigation environment. Based on the offset distance and the corresponding weights, the module determines the compliance of the ship's track during navigation by logical superposition.

[0015] Preferably, the tracking and monitoring control unit includes: The derivation relationship establishment module is used to acquire and classify temperature data and navigation conditions within the expected navigation period according to temperature gradient and traffic complexity, and establish a consumption rate derivation relationship based on temperature data and navigation conditions. The supply duration estimation module is used to obtain the consumption rate of ship hydraulic oil by using the consumption rate derivation relationship, and to estimate the supply duration of the ship under the consumption rate condition based on the total amount of ship hydraulic oil carried and the adjusted sailing speed. The sailing speed adaptation assessment module is used to determine the adaptation degree of the adjusted sailing speed based on the supply time, and to identify the deficiencies of the adjusted sailing speed based on the adaptation degree. It adopts segmented verification and linkage adjustment logic to optimize the sailing speed to ensure the supply quality of ship hydraulic oil.

[0016] The beneficial effects of this invention are as follows: 1. This invention, through the fusion processing of multi-source signals and prediction of target trajectories, can provide a scientific and reliable basis for decision-making regarding ship navigation status. Furthermore, based on the calculation results of the offset distance between the ship's operating trajectory and the route to be navigated, the compliance of the navigation track can be evaluated, and illegal navigation behavior of ships can be detected in a timely manner, ensuring maritime traffic order and navigation safety. In addition, by combining environmental conditions to predict the expected time of arrival of ships at maritime service areas, the planning and timeliness of ship navigation are improved. This invention achieves multi-dimensional and high-precision ship prevention, control, tracking and monitoring, providing comprehensive technical support and guarantee for ship navigation safety, traffic management, and improved operational efficiency.

[0017] 2. This invention takes full-process automated prevention and control as its core objective, reducing the uncertainty of manual intervention. On the one hand, it relies on multi-source information fusion technology to break down the data silos of AIS, radar and video surveillance systems. On the other hand, it builds a software and hardware collaborative linkage mechanism to integrate signal processing, target tracking and other links into an automated process, ultimately achieving accurate identification, rapid response and efficient prevention and control of ship navigation status, and comprehensively improving the stability and reliability of the monitoring system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of a ship automatic prevention, control, tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention. Figure 2 This is a flowchart of the operation of an automatic ship control, tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention. Figure 3 This is an operational trajectory diagram of a ship automatic prevention, control, tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the optimal magnification in an automatic ship control and tracking monitoring system based on the fusion of AIS, radar, and video signals according to an embodiment of the present invention. Figure 5 This is one of the working diagrams of ship identification AI model discovery and tracking in the ship automatic prevention and control tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention; Figure 6 This is the second diagram of the ship identification AI model discovery and tracking operation in the ship automatic prevention and control tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention.

[0020] In the picture: 1. Video signal capture unit; 2. Signal fusion trajectory prediction unit; 3. Flight path compliance assessment unit; 4. Arrival time prediction unit; 5. Tracking and monitoring control unit. Detailed Implementation

[0021] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0022] According to an embodiment of the present invention, an automatic ship control and tracking monitoring system based on the fusion of AIS, radar, and video signals is provided.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figures 1-2 As shown, the ship automatic prevention, control, tracking and monitoring system based on the fusion of AIS, radar and video signals according to an embodiment of the present invention includes: The video signal capture unit 1 is used to adjust the tracking state of the camera capture device according to the latitude, longitude and angle adjustment technology of the camera capture device, and to capture the video signal of the ship during the navigation process using the camera capture device after the tracking state adjustment.

[0024] In one embodiment, a latitude, longitude, and angle adjustment strategy is configured for the camera capture device. The installation coordinates of the camera capture device are preset (based on the WGS84 coordinate system), and the angle adjustment accuracy is calibrated to ensure that the camera capture device can automatically adjust the horizontal and vertical turning angles and zoom levels according to the position changes of the target ship. The calibrated camera capture device is then deployed in areas requiring monitoring and control, such as port terminals and key nodes in waterways. After deployment, the tracking status calibration of the camera capture device is initiated. The response speed of the latitude, longitude, and angle adjustment technology is adjusted using a preset benchmark target (such as a fixed moored ship) to ensure that the camera capture device can complete the tracking status adjustment within 500ms when the target ship moves. Using the camera capture device after the tracking status adjustment, the video signal of the ship during navigation is continuously captured. The capture frequency is set to 25 frames / second to ensure the continuity of the video signal. At the same time, the captured video signal is transmitted in real time to the signal fusion trajectory prediction unit 2 to provide visual data support for subsequent signal fusion and trajectory prediction.

[0025] The signal fusion trajectory prediction unit 2 interacts with the video signal capture unit 1 to fuse ship digital signals, radar digital signals and video signals, and predict the ship's trajectory within the target time period based on the signal fusion results.

[0026] In one embodiment, the signal fusion trajectory prediction unit 2 includes: The digital signal acquisition module is used to deploy the Automatic Identification System (AIS) and radar detector. The AIS device is set to send a digital signal of the ship every 30 seconds, and the radar detector is set to send a digital radar signal every 10 seconds. The AIS receives broadcast signals sent by the ship itself to acquire digital signals of the ship, such as ship name, MMSI code, real-time latitude and longitude, speed, and heading. The radar detector transmits electromagnetic waves and receives reflected signals to acquire radar digital signals of the ship, such as distance, bearing, and speed. At the same time, it synchronously receives video signals transmitted by the video signal capture unit 1 to complete the initial acquisition of the three types of signals. The signal fusion output module is used to calculate the distance between the ship's digital signal, radar digital signal and video signal, determine the corresponding trust function, and perform signal fusion processing by combining fuzzy clustering technology to output fused signal data. The navigation trajectory prediction module is used to combine fused signal data with ship route maps to output a set of candidate trajectory endpoints, and generate a heat map based on the set of candidate trajectory endpoints to predict the ship's operating trajectory within the target time period.

[0027] In one embodiment, the distances between the ship's digital signal, radar digital signal, and video signal are calculated, and the corresponding trust functions are determined. Then, fuzzy clustering technology is used to perform signal fusion processing, and the output fused signal data includes: The acquired ship digital signals, radar digital signals, and video signals are preprocessed. First, outliers (such as latitude and longitude values ​​outside the navigation channel in AIS signals, abnormally high speed values ​​in radar signals, and unclear frame data in video signals) are removed using the 3σ criterion. Simultaneously, a min-max normalization method is used to normalize the numerical range of the three signals to the [0,1] interval, eliminating fusion errors caused by differences in signal magnitude. After preprocessing, the distance between any two signals is calculated: Euclidean distance between ship digital signals and radar digital signals, cosine distance between ship digital signals and video signals, and Manhattan distance between radar digital signals and video signals. A 3×3 distance matrix is ​​constructed based on the calculation results. Based on this distance matrix, the element similarity of the three signals is calculated using the Pearson correlation coefficient. After normalizing the element similarity, the following calculations are performed. The trust function between any two signal classes is used. It's important to note that the trust function ranges from [0,1]. A larger value indicates higher consistency between the two signal classes. The preset trust threshold is 0.6. Signal data with a trust function below 0.6 are removed. For example, if the trust function between a video signal and an AIS signal is 0.55, that frame of video signal data is removed. The mean of the remaining signal data is calculated, and outlier data (data with subtle differences among the three signal classes that were not removed) is output. Principal component analysis (PCA) is used to process the outlier data, extracting the principal components and retaining those with a cumulative contribution rate ≥95%. A similarity matrix is ​​calculated based on the processing results to capture the relationships between outlier data and reduce data redundancy. Based on the similarity matrix, a cluster radius of 0.3 is set, and the K-means fuzzy clustering algorithm is used to perform iterative optimal solution selection. The preset number of iterations is 50. When the iteration error is ≤10... -4 The iteration stops when the optimal solution is determined. The weight vector is calculated using the optimal solution, and a membership matrix is ​​generated. The heterogeneous data is then fused using the membership matrix, and finally the fused signal data is output.

[0028] In one embodiment, the process of using fused signal data and a ship route map to output a set of candidate trajectory endpoints and generating a heat map based on the candidate trajectory endpoint set to predict the ship's trajectory within a target time period includes: using fused signal data and ship route map information, employing graph neural network technology to jointly encode the ship's trajectory information and ship route map information to obtain a scene encoding vector; obtaining a set of candidate trajectory points for the ship within the navigation area based on the scene encoding vector, and analyzing the probability value of each candidate trajectory point in the candidate trajectory point set within the target time period using an attention mechanism; generating a heat map based on the probability values ​​of each candidate trajectory point to present the probability of the ship appearing within the navigation area within the target time period, and predicting the ship's trajectory within the target time period based on the appearance probability and trajectory completion processing method, thereby determining the ship's trajectory.

[0029] In one embodiment, based on fused signal data and combined with ship route map information, graph neural network technology is used to jointly encode the ship's trajectory information and ship route map information to obtain a scene encoding vector. This includes: splitting the fused signal data into trajectory vectors connected end-to-end according to time sequence, while defining the ship route map information as map vectors. The map vectors describe the geographical features of ship navigation. The trajectory vectors and map vectors are used as graph nodes, and adjacent vectors are used as edges to pass node information and extract local features, resulting in the encoding features of the ship trajectory local map and the map local map; the encoding features are used as input to perform global information fusion through graph neural network to construct a global scene map containing the ship's historical trajectory and navigation environment, resulting in a two-dimensional scene encoding feature matrix; the fused signal data is embedded into standardized coordinates to obtain an embedding vector, which represents the ship's coordinate encoding information at different times. The embedding vector is then sequentially input into the encoder according to each time to extract the temporal features of the ship's motion state; the temporal features output by the encoder are finally input into a fully connected layer containing an activation function to obtain the encoding vector of the ship's motion state.

[0030] In one embodiment, obtaining a set of candidate trajectory points for the ship within the navigation area based on the scene encoding vector, and analyzing the probability value of each candidate trajectory point in the candidate trajectory point set within the target time period using an attention mechanism includes: determining the navigation area based on the ship's route map information; within the navigation area, combining the target prediction time period with the ship's navigation speed scene encoding map to determine the ship's trajectory points within the navigation area; integrating candidate trajectory points using a dense sampling method, and removing candidate points whose distance is higher than a threshold and which conflict with obstacles in the navigation area to obtain the candidate trajectory point set; fusing the two-dimensional scene encoding feature matrix with the encoding vector to perform two-dimensional scene encoding... The ship-related encoded vectors in the feature matrix are superimposed, and the irrelevant row vectors are filled with zero vectors to obtain the fusion matrix. Based on the fusion matrix and the two-dimensional scene encoded feature matrix, a comprehensive encoding matrix containing ship navigation scene and operation status information is determined. The comprehensive encoding matrix and the candidate trajectory point set are mapped to a query matrix, a key matrix, and a value matrix using a projection matrix. Based on the query matrix, key matrix, and value matrix, the attention distribution of the association weights between the candidate trajectory points and the encoded vectors is calculated using a scaling point attention mechanism. Based on the attention distribution, the trajectory driving candidate points after fusing attention features are obtained and input to the fully connected layer to output the probability value of the trajectory driving candidate points in the target time period.

[0031] It needs to be explained that the core implementation process of the navigation trajectory prediction module is to acquire ship route map information (using the WGS84 coordinate system, including information such as channel boundaries, obstacle positions, and water depth). Based on the fused signal data, combined with the ship route map information, graph neural network (GNN) technology is used to jointly encode the ship's trajectory information and the ship route map information to obtain a scene encoding vector. This is achieved by splitting the fused signal data into contiguous trajectory vectors (e.g., t0-t) according to time sequence (one data point every 10 seconds). 10 t 10 -t 20 t 20 -t 30 The corresponding trajectory vector), and the ship's route map information is defined as a map vector. The map vector describes the ship's navigation geographical features (such as channel width, obstacle coordinates, turning nodes, etc.). The trajectory vector and the map vector are used as graph nodes, and adjacent vectors (such as t0-t) 10 With t 10 -t 20Using the trajectory vector (and the map vectors of adjacent regions) as edges, a graph convolutional layer is used for node information transfer and local feature extraction. The convolutional kernel size is set to 3×3, and the activation function is ReLU. This yields the encoded features of the ship's trajectory local map and the map local map. These encoded features are then used as input for global information fusion through a fully connected layer of a graph neural network. The number of hidden layer neurons is set to 128, the number of iterations is 100, and the learning rate is 0.001. This constructs a global scene graph containing the ship's historical trajectory and navigation environment, resulting in a 256×256 two-dimensional scene encoding feature matrix. The fused signal data is then embedded into standardized coordinates within the range [-1,1], yielding an embedding vector of dimension 1×64. This embedding vector represents the ship's coordinate encoding information at different times, such as time t0. The embedding vector represents the ship's latitude, longitude, speed, and heading at a given moment. This embedding vector is then sequentially input into the encoder, which can be an LSTM encoder with one input layer, two hidden layers, and one output layer. The hidden layer has 64 neurons. This process extracts temporal features of the ship's motion state, such as speed trends and turning patterns. The encoder's final output, a 1×128-dimensional temporal feature vector, is then input into a fully connected layer containing a sigmoid activation function, resulting in a 1×64-dimensional encoded vector of the ship's motion state. This completes the joint encoding, yielding a scene encoding vector (composed of a two-dimensional scene encoding feature matrix and the ship motion state encoding vector). Furthermore, the ship's trajectory information and navigation environment information can be fused and encoded, transforming them into a feature vector suitable for trajectory prediction.

[0032] After obtaining the scene encoding vector, a set of candidate trajectory points for the ship within the navigation area is acquired based on this vector. An attention mechanism is then used to analyze the probability value of each candidate trajectory point within the target time period. The navigation area is determined based on the ship's route map information. In this embodiment, the navigation area is assumed to be a rectangular region between 30°15′00″ and 30°16′00″ N and 120°05′00″ and 120°06′00″ E, with a channel width of 100 meters. Two obstacles exist within this region: obstacle 1 has coordinates of 30°15′30″ N and 120°05′20″ E; obstacle 2 has coordinates of 30°15′45″ N and 120°05′50″ E. The target prediction time period is the next 10 seconds (i.e., t). 30 -t 40(Time period), the ship's current speed is 10 knots. Combining the scene encoding vector, a graph neural network model is used to simulate the ship's movement trend within the navigation area, determining the ship's trajectory points within the navigation area. Then, a dense sampling method is used, with a sampling interval of 1 meter to integrate trajectory candidate points, collecting a total of 80 candidate points. Candidate points whose distances are higher than a threshold (the threshold is set to 5 meters, i.e., distance from the channel boundary > 5 meters, distance from the obstacle > 5 meters) and conflict with obstacles in the navigation area are removed. For example, candidate point coordinates are 30°15′31″N, 120°05′20″E, and the distance from obstacle 1 is 1 meter, belonging to conflict candidate points and being removed. Finally, 56 candidate points are obtained, forming a candidate trajectory point set; the two-dimensional scene encoding feature matrix (256×256) is then used. The system is fused with the ship motion state encoding vector (1×64). The ship-related encoding vectors in the two-dimensional scene encoding feature matrix, such as the encoding vectors corresponding to the ship's current position and historical trajectory, are superimposed. Irrelevant row vectors, such as the encoding vectors corresponding to the outer area of ​​the waterway, are filled with zero vectors to obtain a 256×256 fusion matrix. Based on the fusion matrix and the two-dimensional scene encoding feature matrix, a comprehensive encoding matrix (256×256) containing ship navigation scene and operation status information is calculated by matrix multiplication. Then, the comprehensive encoding matrix is ​​combined with the candidate trajectory point set, i.e., 56 candidate points, using a projection matrix (256×64). Each candidate point corresponds to a 1×64 vector mapping to a query matrix (256×64), a key matrix (256×64), and a value matrix (256×64).

[0033] Based on the query matrix, key matrix, and value matrix, a scaling point attention mechanism is used to calculate the attention distribution of the association weights between candidate trajectory points and encoding vectors. Specifically, the product of the transpose of the query matrix and the key matrix is ​​first calculated to obtain the attention score matrix. Then, the attention score matrix is ​​divided by the square root of the dimension of the key matrix for scaling to avoid gradient vanishing due to excessively high scores. The scaled attention score matrix is ​​normalized using the softmax function to obtain the attention distribution, which is the association weight of each candidate trajectory point. Based on the attention distribution, the value matrix and the attention distribution are weighted and summed to obtain the trajectory driving candidate points after fusing attention features. This is then input into a fully connected layer with 64 hidden layer neurons and ReLU activation function. The output is the probability value of each trajectory driving candidate point in the target time period, with the probability value ranging from [0,1]. The higher the probability value, the greater the possibility that the ship will arrive at the candidate point in the target time period. This allows for the selection of reasonable candidate trajectory points and the quantification of the arrival probability of each candidate point.

[0034] After obtaining the probability values ​​of each candidate trajectory point, a heatmap generation tool (such as Matplotlib) is used with the navigation area map as the background to map the probability value of each candidate trajectory point to a heatmap value. A probability value of 0 corresponds to blue, a probability value of 0.5 corresponds to yellow, and a probability value of 1 corresponds to red, generating a heatmap. In the heatmap, red areas indicate a very high probability of ship occurrence (≥0.8), yellow areas indicate a medium probability (0.3-0.8), and blue areas indicate a very low probability (≤0.3). Based on the heatmap, candidate trajectory points with a probability value ≥0.7 are selected. In this embodiment, 12 are selected. Linear interpolation is used to complete the trajectory of the candidate trajectory points, filling the gaps between them. The interpolation interval is 0.5 meters, resulting in continuous trajectory segments. The continuity and rationality of the trajectory segments are judged by combining historical ship trajectories, and discontinuous trajectory segments are eliminated, such as those showing sudden turns or abrupt speed changes. Finally, the target time period (t) is determined. 30 -t 40 The ship's trajectory, assuming the predicted trajectory is: from t 30 Departing from coordinates (30°15′10″N, 120°05′10″E), sailing at a constant speed of 10 knots, with the bow pointing 90° (due east), t 40 The trajectory arrives at the coordinates (30°15′10″N, 120°05′40″E). The trajectory is located within the waterway, and the distance to obstacles is greater than 5 meters, with no risk of collision. Therefore, based on the probability value of the candidate trajectory points, the ship's operating trajectory within the target time period can be predicted, providing a decision-making basis for automatic ship control and tracking.

[0035] It needs to be explained that AIS digital signals, radar digital signals, and video capture signals are fused together to form a single fused digital signal. This signal possesses the attributes of AIS, radar, and vision, while also offering higher real-time performance and accuracy, and a certain degree of predictive ability, capable of predicting the trajectory of a ship target within the next few seconds. Figure 3 As shown, the fused digital signal is based on the AIS digital signal, merging the radar digital signal and the visual capture signal. Simultaneously, based on the latitude and longitude in the AIS digital signal and the ship's heading and speed in the radar digital signal, the target's latitude and longitude one second in the future is simulated and placed into a new signal channel. Secondly, since the visual capture signal is not always present, the fused signal allows for the visual signal to be omitted. Under the premise of satisfying the visual capture signal, the visual capture data is fused in through a weighting algorithm (as shown in Formula 1). After the next AIS digital signal and radar digital signal are received, the simulated latitude and longitude, ship speed, and heading data in the fused signal are compared with the data in the actual signal to calibrate the predicted signal data and improve the accuracy of the simulated signal.

[0036] Formula 1: Weighting algorithm for multi-parameter signal fusion of ship targets; Latitude and longitude distance calculation (weight 70%): L=R×arccos[sinφ1×sinφ2+cosφ1×cosφ2×cos(Δλ)]; In the formula, L (meters) represents the actual distance between two points (core output, error < 500m); R (6371000 meters) represents the average radius of the Earth (WGS84 coordinate system constant); φ1 / φ2 (radians) represents the latitude of the target / reference point (degrees, minutes, seconds converted to radians, positive for North latitude and negative for South latitude); λ1 / λ2 (radians) represents the longitude of the target / reference point (degrees, minutes, seconds converted to radians, positive for East longitude and negative for West longitude); Δλ (radians) represents the difference in longitude, i.e., |λ2-λ1|.

[0037] Heading angle calculation (weight 20%): When the result is ≥0, θ = arctan2(Δλ × cosφ) m ,Δφ)×(180 / π); When the result is less than 0, θ = arctan2(Δλ × cosφ) m ,Δφ)×(180 / π)+360; In the formula, θ (degrees) represents the target heading angle (core output, error < 15°), Δφ (radians) represents the latitude difference φ1-φ2 ​​(φ2 / φ1 is the latitude at time t2 / t1), Δλ (radians) represents the longitude difference λ2-λ1 (λ2 / λ1 is the longitude at time t2 / t1), and φ m (Radians) represents the average latitude (φ1+φ2) / 2, corrected for the curvature of the Earth.

[0038] Speed ​​calculation (weight 10%): V = (L × 3600) / (1852 × Δt); In the formula, V (knots) represents the sailing speed (core output, error < 3 knots), L (meters) represents the distance traveled from time t1 to t2 (obtained from the latitude and longitude distance formula), Δt (seconds) represents the time difference t2-t1, and 1852 (meters / nautical miles) and 3600 (seconds / hours) represent unit conversion factors.

[0039] Calculation of total weight percentage (must be >70%): S = 70% × x L +20%×x θ +10%×x V (Requirement: S > 70%) In the formula, S (0~100%) represents the total weight percentage (core judgment indicator), x L x represents the distance error compliance variable (1 = error < 500m, 0 = otherwise),θ x represents the variable indicating whether the heading error meets the standard (1 = error < 15°, 0 = otherwise), V The variable representing the speed error meeting the standard (1 = error < 3 sections, 0 = otherwise) corresponds to a weight of 70%, 20%, and 10%, respectively.

[0040] Meanwhile, the three-source tracking algorithm uses the fusion signals of AIS and radar, combined with the latitude and longitude of the camera itself, to guide the monitoring camera to turn and track in real time through the AI ​​target recognition and target trajectory prediction algorithms of the large model, and uses angle algorithms to assist in tracking. At the same time, it ensures that the target is always in the center of the monitoring camera's image through preset optimal field of view, and displays the best observation size of the target through camera zoom.

[0041] The camera itself has orientation calibration, using the WGS84 coordinate system as the default coordinate system and the default true north direction as the 0° angle. By acquiring the target's latitude and longitude coordinates and the camera's coordinates in real time, the target distance L is calculated using the short-range formula (as shown in Formula 2), and the angle θ is calculated using the angle formula (as shown in Formula 3). The camera is then directed to turn and track the target in real time.

[0042] Formula 2: The short distance formula is D=K×M×(θ–Δθ), unit: meters; in the constants, K=6378137m, e=1 / 298.257223563; Input validation is φA, φB latitude (radians); λA, λB longitude (radians); in the auxiliary parameter calculation, the latitude correction value is UA=arctan[(1-e)×tanφA], UB=arctan[(1-e)×tanφB]; the longitude difference is Δλ=λB-λA, and iterative calculation is performed (until |λnew-λprev|<1×10-10).

[0043] The calculation process related to chord length is as follows: sinθ=(cosUB×sinλ)2+(cosUA×sinUB−sinUA×cosUB×cosλ)2cosθ=sinUA×sinUB+cosUA×cosUB×cosλθ=arctan2(sinθ,cosθ); Azimuth related: sinβ=(cosUA×cosUB×sinλ) / sinθcos2β=1-sin2β; Iterative update of longitude difference: λnew=Δλ+(1-e / 16)×e×sinβ×θ; The auxiliary coefficient in the final distance calculation is M=1+(e2×cos2β) / 10000; the correction term is Δθ=0.001×sinθ; the final distance is D=K×M×(θ-Δθ).

[0044] Formula 3 is calculated as follows: ; In the formula, Δλ=λ–λ0 represents the difference between the longitude of the target and the longitude of the camera; φ0 and λ0 represent the latitude and longitude coordinates of the camera; φ and λ represent the latitude and longitude coordinates of the target; and θ represents the azimuth in radians. After converting it to an angle, it needs to be standardized to the range of 0° to 360°.

[0045] Once the camera detects a target, AI performs target recognition and calculates the target's distance, direction of travel, and speed. This data is then sent to the signal fusion channel. Simultaneously, based on the preset optimal field of view and the field of view offset β and β′ between the target recognition field and the target's field of view, the system automatically calculates the precise horizontal and vertical angles the camera should rotate. This, combined with the horizontal and vertical center lines of the image, ensures the target is centered in the frame. Furthermore, based on the field of view deviation α and α′ between the optimal field of view and the target recognition field, the system automatically adjusts the camera's magnification to maintain the target at its optimal observation size. Figure 4 As shown.

[0046] AI-based ship target recognition technology is designed for real-world ship targets. It employs an image recognition AI model for specialized training, incorporating both daytime and nighttime characteristics of the ship during the training process. During the day, image data is collected using cameras equipped with visible light sensors, while at night, relevant images are acquired using cameras equipped with thermal imaging sensors. This process ultimately constructs a dedicated ship recognition AI model, which can be used to assist in the detection and tracking of ship targets. Figure 5 and Figure 6 As shown.

[0047] Track compliance assessment unit 3 is used to obtain the route to be navigated by the ship and calculate the offset distance between the ship's operating trajectory and the route to be navigated, and to assess the ship's track compliance during navigation based on the offset distance.

[0048] In one embodiment, the flight track compliance assessment unit 3 includes: The route division module is used to obtain the route to be navigated by the ship, divide the route to be navigated by the ship into three stages according to the route status, and determine the anchoring feature points of each segment of the route to be navigated by the ship. The three stages include the mandatory segment, the adjustment segment, and the warning segment. The offset distance definition module is used to synchronize and align the ship's trajectory with the segmented results of the ship's route according to time nodes, and define the offset distance between the ship's trajectory and the route to be navigated based on anchor feature points and segment comparison logic. The navigation compliance judgment module is used to assign weights to the offset distance of each segment of the ship's route according to the navigation environment. Based on the offset distance and the corresponding weights, the module determines the compliance of the ship's track during navigation by logical superposition.

[0049] It needs to be explained that the core objective of the Track Compliance Assessment Unit 3 is to scientifically assess the compliance of a ship's track by accurately defining the deviation distance between the ship's actual operating trajectory and the intended navigation route, and dynamically assigning weights based on the navigation environment. This allows for the timely detection of potential track deviations and provides a basis for compliance decision-making in ship prevention and control monitoring. The unit obtains the ship's intended navigation route through the ship dispatcher, electronic chart system, or the route plan reported by the ship itself. The intended navigation route must include basic information such as a complete sequence of latitude and longitude coordinates, total route length, estimated travel time, and key nodes (e.g., turning points, channel entrances / exits, restricted area boundaries). Simultaneously, it acquires the corresponding navigation environment information (e.g., channel width, water depth, weather conditions, traffic density), and divides the intended navigation route into three stages according to its status: a mandatory stage, an adjustment stage, and a warning stage. The division is based on channel management requirements, the complexity of the navigation environment, and... The operational requirements for vessels are defined, and the specific division logic is as follows: The mandatory section is a section where the waterway management department explicitly stipulates that vessels must strictly adhere to the designated route and are not allowed to deviate beyond the prescribed range. Examples include narrow channels, channels under bridges, and core channels surrounding prohibited areas. The navigation environment in this section is complex, with extremely low tolerance for error; deviation can easily lead to collisions, groundings, and other accidents. The adjustment section is a section where vessels can make minor adjustments within a specified deviation range, such as wide channels and sections with low traffic density. The navigation environment in this section is relatively relaxed, allowing vessels to make reasonable deviations based on actual navigation conditions, such as avoiding other vessels or adjusting their course, without needing to strictly adhere to the route. The warning section is between the mandatory and adjustment sections, where timely warnings and course adjustments are required when vessel deviation reaches a preset warning threshold (channel turning transition sections and sections with moderate traffic density). Vessel deviation in this section requires close monitoring to proactively avoid the risk of escalating deviations.

[0050] After the route is divided, anchorage feature points for each segment of the vessel's navigation route are determined. These anchorage feature points serve as the benchmark for subsequent offset distance calculations. The selection criteria are the starting point, ending point, turning point, key nodes, and coordinate points at regular intervals, based on the segment length. Forced segment intervals are denser, warning segments are less dense, and adjustable segment intervals can be more relaxed. Anchorage feature points must be clearly marked with their latitude and longitude coordinates, feature point type (starting point, ending point, turning point, etc.), and the attributes of the corresponding segment. This ensures that subsequent offset distance determination has a clear benchmark. By scientifically segmenting the vessel's navigation route, clarifying the navigation requirements of each segment, and selecting precise anchorage feature points, a foundation is laid for subsequent offset distance determination and compliance assessment.

[0051] After completing the route segmentation and determining the anchoring feature points, based on the ship's operational trajectory data and the segmentation results of the route to be navigated (start and end times, latitude and longitude ranges, and anchoring feature point coordinates for each segment), the ship's operational trajectory and the segmentation results of the route to be navigated are synchronized and aligned according to time nodes. The core logic of the synchronization and alignment is: based on the expected sailing time of the route to be navigated, each segment of the route to be navigated is mapped to a specific time interval (e.g., the expected sailing time of a certain mandatory segment is t). 40 -t 100 The corresponding time point is t 40 t 50 、…、t 100 The process involves comparing the trajectory points of the ship's actual operating trajectory within the corresponding time interval with the planned navigation route. This ensures that each actual trajectory point corresponds to a specific segment of the planned navigation route, avoiding offset errors caused by time asynchrony. After synchronization and alignment, based on anchoring feature points and a pre-defined segmentation comparison logic, the offset distance between the ship's operating trajectory and the corresponding planned navigation route is defined. The segmentation comparison logic employs differentiated definition methods for three different routes. The specific logic and implementation process include: when the segmentation result of the planned navigation route is a mandatory segment, using the anchoring feature points of that segment as a reference, a baseline for the mandatory segment is fitted using linear interpolation. This baseline represents the standard route that the ship must follow. The baseline is a continuous line segment connecting all anchoring feature points of that segment. The offset distance in the ship's actual operating trajectory is then calculated. The straight-line offset distance between the trajectory point at each time node and the baseline is calculated using the point-to-line distance formula. This distance is the offset distance of the forced segment, which is not allowed to have large offsets. The offset distance calculation must be accurate to 0.1 meters. When the segmentation result of the route to be navigated is an adjustment segment, the preset offset threshold range is used as a reference to determine whether the ship's trajectory is within the allowable offset range. First, based on factors such as the channel width and navigation environment of the adjustment segment, the preset offset threshold range is determined (e.g., the allowable offset range is -10 meters to +10 meters, where a negative value indicates that the trajectory is to the left of the baseline and a positive value indicates that the trajectory is to the right of the baseline). This threshold range can be dynamically adjusted according to the actual navigation environment (e.g., when the channel width increases, the threshold range can be appropriately widened).

[0052] Calculate the straight-line offset distance between the ship's actual trajectory point and the baseline of the adjustment section, fitted by the anchoring feature points of that section. Determine if this offset distance is within a preset offset threshold range. If it is within the range, it is defined as an allowable offset, and the offset distance is recorded as the actual calculated value. If it exceeds the range, it is defined as exceeding the allowable range, and the offset distance is recorded as the difference between the actual calculated value and the threshold (e.g., if the upper limit of the threshold is 10 meters and the actual offset distance is 12 meters, then the excess offset distance is recorded as 2 meters). When the segment result of the route to be navigated is a warning segment, the offset distance and direction between the ship's trajectory and the warning critical point are defined based on the warning critical point. The warning critical point is set according to the navigation risk of the warning segment and is divided into a left-side warning critical point and a right-side warning critical point. The critical point (i.e., a line segment parallel to the baseline at a certain distance to the left and right) is determined by first calculating the straight-line offset distance between the actual trajectory point of the vessel and the baseline, then determining whether the trajectory point is located to the left or right of the baseline, and calculating the offset distance between the trajectory point and the corresponding warning critical point (e.g., if the trajectory point is to the right of the baseline, calculate its distance to the right warning critical point; if the trajectory point has exceeded the warning critical point, the distance is positive; if it has not exceeded it, the distance is negative). Simultaneously, the offset direction (left / right) is recorded to provide directional basis for subsequent compliance judgments and warnings. This allows for precise definition of the offset distance between the actual vessel trajectory and each segment of the planned route, clarifying the degree and direction of the offset, and providing core data support for compliance judgments.

[0053] After defining the offset distances for each segment, real-time environmental information of the vessel's current navigation is obtained, including meteorological conditions (wind force, visibility), channel conditions (channel width, water depth, traffic density), navigation time (day / night), and special control requirements (such as temporary channel control, adjustment of restricted areas, etc.). Based on this navigation environment information, corresponding weights are assigned to the offset distances of each segment (mandatory segment, adjustment segment, warning segment) of the route to be navigated. The core logic of weight assignment is: the more complex and risky the navigation environment, the higher the weight of the segment, that is, the greater the impact of offset distance on compliance. The specific weight assignment rules are set in combination with the actual navigation scenario, and can also be dynamically adjusted through algorithms to ensure the rationality of the weights. Based on the offset distances defined for each segment and the corresponding assigned weights, the overall track compliance score of the vessel is calculated through logical superposition. Then, based on the preset compliance threshold, the track compliance level of the vessel during navigation is determined, and a compliance assessment report and early warning prompts are generated (if non-compliance or warning situations exist). The specific algorithm for weight assignment is hierarchical. The Analysis of Hierarchical Method (AHP) algorithm quantifies the impact of various navigation environment factors on the weights of each route segment by constructing a judgment matrix. First, it determines the target for weight assignment (to assign reasonable weights to the offset distances of each route segment, improving the accuracy of compliance judgments) and constructs a hierarchical structure. The target layer assigns weights, the criterion layer represents the navigation environment factors affecting the weights (including four factors: weather conditions, channel conditions, traffic density, and navigation time), and the option layer represents the routes to be weighted (mandatory, adjustment, and warning segments). Through expert scoring, the importance of each factor in the criterion layer is compared pairwise to construct a criterion layer judgment matrix. Simultaneously, the importance of each route segment in the option layer under each criterion layer factor is compared pairwise to construct an option layer judgment matrix. The eigenvalues ​​and eigenvectors of each judgment matrix are calculated, and a consistency check is performed (the judgment matrix is ​​valid when the consistency check index CR < 0.1). If the check passes, the weights of each criterion layer factor and the weights of each option layer route segment under each criterion layer factor are determined based on the eigenvectors. Finally, through hierarchical sorting, the final weights of the offset distances of each route segment are calculated.

[0054] Specifically, offset distance deduction rules are set for each road segment. The greater the offset distance, the more points are deducted. For road segments with higher weights, the same offset distance results in more deductions. The deduction score for each road segment is calculated as follows: Deduction score = Offset distance × Corresponding road segment weight × Deduction coefficient. The deduction coefficient is a preset constant set according to the road segment type, with the highest deduction coefficient for mandatory segments and the lowest for adjustment segments. The overall compliance score is calculated as follows: Compliance score = 100 - Sum of deduction scores for each road segment. The compliance score ranges from 0 to 100. Simultaneously, a compliance threshold is preset, divided into three levels. Level: Excellent (90-100 points, track fully compliant, no deviation or slight deviation, no adjustment required), Qualified (70-89 points, track basically compliant, slight deviation exists but does not exceed the warning range, continuous monitoring required), Non-compliant (0-69 points, track seriously deviates, exceeds the allowable range, immediate warning required and vessel must adjust course); If the vessel's trajectory exceeds the allowable deviation range of the mandatory segment, the adjustment segment seriously exceeds the threshold range, or the warning segment exceeds the warning threshold and is not adjusted in time, it will be directly judged as non-compliant.

[0055] Assuming the total length of the route to be navigated is 2000 meters and the estimated navigation time is 200 seconds (t0-t... 200 The route traverses a port access channel. Considering channel management requirements and navigation environment, it is divided into three phases using a route segmentation module. The specific segmentation results are as follows: Forced Segment (t0-t60): 600 meters long, a narrow channel (20 meters wide), with high traffic density (3 other vessels navigating nearby). Weather conditions are clear, wind force 3, and visibility 10 kilometers. This section is a channel under a bridge, and significant deviations are strictly prohibited. After segmentation, five anchorage feature points were determined for this segment: P1 (30°15′05″N, 120°05′05″E, t0 time, starting point), P2 (30°15′07″N, 120°05′15″E, t60 time, starting point, t7′07″N, t60 time, starting point ... 20 (Time), P3 (30°15′09″N, 120°05′25″E, t) 40 Time), P4 (30°15′11″N, 120°05′35″E, t 50 Time), P5 (30°15′13″N, 120°05′45″E, t 60 (Time, End Point), anchor feature point interval is 150 meters (forced segment interval densification).

[0056] Among them, the adjustment segment (t) 60 -t 140 The section is 800 meters long, a wide channel (50 meters wide), with low traffic density (no other vessels navigating nearby). Weather conditions remain unchanged. Minor adjustments by vessels are permitted in this section. Four anchoring feature points have been identified, namely P5 (the endpoint, t...). 60Time), P6 (30°15′15″N, 120°05′55″E, t 80 Time), P7 (30°15′17″N, 120°06′05″E, t 100 Time), P8 (30°15′19″N, 120°06′15″E, t 140 Time, End Point), anchor feature point interval is 200 meters; warning section (t 140 -t 200 The section is 600 meters long and serves as a turning transition section (turning angle 30°). Traffic density is moderate (one other vessel is navigating nearby). Weather conditions remain unchanged. This section requires close monitoring of any deviations. Four anchoring feature points have been identified: P8 (end point, t...). 140 Time), P9 (30°15′21″N, 120°06′20″E, t 160 Time), P10 (30°15′23″N, 120°06′25″E, t 180 Time), P11 (30°15′25″N, 120°06′30″E, t 200 The time and destination are set, and the anchor feature points are spaced 150 meters apart. After the anchor feature points are determined, the latitude and longitude coordinates and corresponding time nodes of each feature point are recorded simultaneously to complete the route division and anchor feature point determination.

[0057] The ship trajectory data output by the signal fusion trajectory prediction unit 2 is obtained (time nodes are t0, t1, t2). 10 t 20 、…、t 200 (One trajectory point every 10 seconds) The actual trajectory is synchronized and aligned with the three road segments according to the time nodes, i.e., t0-t 60 The actual trajectory point corresponds to the forced segment, t 60 -t 140 The actual trajectory point corresponds to the adjustment segment, t 140 -t 200 The actual trajectory points correspond to the warning segments. After synchronous alignment, the offset distance of each segment is defined based on segment comparison logic: the baseline of the forced segment is fitted with the anchor feature points of P1-P5 using linear interpolation to obtain a continuous baseline. t is selected. 20 t 40 t 50 The offset of the actual trajectory points at three key time nodes is calculated, t 20The actual trajectory point at time P2′ (30°15′07″N, 120°05′14″E) is converted to Cartesian coordinates (x1=120.0847°, y1=30.2514°), P2 (x2=120.0875°, y2=30.2519°), and P2′ (x=120.0872°, y=30.2519°) (x1=13356200m, y1=3356200m; x2=13356500m, y2=3356250m; x=13356470m, y=3356250m). Substituting these coordinates into the formula for the distance from a point to a straight line, the offset distance d≈4.9 meters is calculated. 40 The actual trajectory point at time P3′ (30°15′09″N, 120°05′26″E) is calculated to be 5.2 meters; t 50 The actual trajectory point at time P4′ (30°15′11″N, 120°05′34″E) has a calculated offset distance of 4.7 meters, meaning the offset distance of the forced segment is between 4.7 and 5.2 meters. The preset offset threshold range for the adjustment segment is -10 meters to +10 meters. The baseline is fitted with anchoring feature points P5-P8, and t is selected. 80 t 100 t 120 The actual trajectory points at three time points, t 80 The actual trajectory point at time P6′ (30°15′15″N, 120°05′57″E) has a calculated offset distance of 6.3 meters (within the threshold range, allowing for offset); t 100 The actual trajectory point at that moment was P7′ (30°15′17″N, 120°06′03″E), with an offset distance of -4.8 meters (leftward offset, within the threshold range); t 120 The actual trajectory point at time P7″ (30°15′17″ N, 120°06′08″ E) has an offset distance of 8.9 meters (within the threshold range). The preset warning threshold for the left side of the warning segment is 5 meters to the left of the baseline, and the preset warning threshold for the right side is 5 meters to the right of the baseline. The baseline is fitted with anchoring feature points P8-P11, and t is selected. 160 t 180 The actual trajectory points at two time points, t 160 The actual trajectory point at time P9′ (30°15′21″N, 120°06′22″E) has a calculated offset distance of 6.1 meters from the baseline (offset to the right), and an offset distance of 6.1 - 5 = 1.1 meters from the right-side warning threshold (exceeding the warning threshold, a positive value). The offset direction is to the right. 180The actual trajectory point at the time is P10′ (30°15′23″N, 120°06′24″E). The calculated offset distance from the baseline is 3.8 meters (offset to the right). The offset distance from the right warning threshold is 3.8-5=-1.2 meters (not exceeding the warning threshold, negative value). The offset direction is to the right, thus completing the definition of the offset distance for each road segment.

[0058] The Analytic Hierarchy Process (AHP) was used to assign weights to each route segment. First, a judgment matrix was constructed. The importance of the four factors in the criterion layer was ranked as follows: waterway conditions > navigation density > meteorological conditions > navigation time period. The criterion layer judgment matrix was constructed, and the eigenvalues ​​and eigenvectors were calculated. The consistency test CR=0.08<0.1, indicating that the judgment matrix was valid. The weights of each factor in the criterion layer were obtained as follows: waterway conditions 0.4, navigation density 0.3, meteorological conditions 0.2, and navigation time period 0.1.

[0059] Construct a scheme-level judgment matrix. The importance ranking of each road segment under each criterion-level factor is: mandatory segment > warning segment > adjustment segment. Calculate the weight of each road segment under each criterion-level factor. Through the overall hierarchical ranking, the final weight of each road segment is obtained as: mandatory segment 0.5, adjustment segment 0.2, and warning segment 0.3.

[0060] The pre-set deduction coefficients for each route segment are: 0.8 for the mandatory segment (most stringent deduction), 0.3 for the adjustment segment, and 0.5 for the warning segment. The compliance thresholds are 90-100 points for excellent, 70-89 points for qualified, and 0-69 points for non-compliant. The deduction values ​​for each segment are calculated as follows: For the mandatory segment, an average deviation distance of 5.0 meters results in a deduction of 2.0 points; for the adjustment segment, an average deviation distance of 6.7 meters results in a deduction of approximately 0.4 points; and for the warning segment, an average deviation distance of 5.0 meters results in a deduction of 0.75 points. The overall compliance score is 96.85 points, which is considered excellent. The vessel's course is deemed fully compliant, requiring no course adjustment. A compliance assessment report is generated, recording the deviation distance, weight, deduction, and compliance level for each segment, and is simultaneously sent to the vessel's bridge and the monitoring center. If we assume t... 50 If the actual deviation of the mandatory segment is 12 meters (exceeding the allowable range of the mandatory segment), the mandatory segment will be penalized with 4.8 points. The overall compliance score is 94.05 points, which is still considered excellent, but a slight warning will be issued to prompt the ship to make minor adjustments to its course.

[0061] If t 50 The actual deviation during the mandatory course segment was 20 meters, resulting in a deduction of 8.0 points. The overall compliance score was 90.85, which is near the "excellent" threshold. Therefore, the warning has been upgraded, and the vessel is advised to immediately adjust its course. 50The actual deviation during the mandatory segment was 30 meters, resulting in a deduction of 12.0 points. The overall compliance score was 86.85, which is considered acceptable. The vessel's deviation will be continuously monitored; if the deviation continues to increase, further warnings will be issued. 50 The actual deviation distance of the mandatory segment was 50 meters, and the deduction score for the mandatory segment was 20.0 points. The overall compliance score was 78.85 points, which is considered qualified. A serious warning should be issued, requiring the vessel to immediately adjust its course to ensure compliance with the track.

[0062] Arrival time prediction unit 4 is used to adjust the ship's speed according to the compliance of the navigation track, and to assess the expected time when the ship will arrive at the water service area based on the adjusted ship speed and the environmental conditions in the navigation area.

[0063] It should be explained that the arrival time prediction unit 4 first receives the complete track compliance assessment result output by the track compliance assessment unit 3. When the ship's track compliance level is excellent (90-100 points), it means that the ship's current track is completely compliant, with no deviation or only a slight deviation, and no course adjustment is required. At this time, the current speed can be maintained, or the speed can be appropriately increased (not exceeding 10% of the current speed) if the navigation environment permits, to ensure navigation efficiency. When the ship's track compliance level is acceptable (70-89 points), it means that the ship's current track is basically compliant, with a slight deviation but not exceeding the warning range. The track status needs to be continuously monitored. At this time, the current speed must be maintained, and the speed must not be increased. If the current speed is too high, it may cause the deviation to worsen, and the speed needs to be slightly reduced. (The speed reduction should be 5%-10% of the current speed) to ensure stable and compliant navigation. When the ship's navigation compliance level is non-compliant (0-69 points), it indicates that the ship's current navigation is seriously deviating from the permitted range. An immediate warning and forced speed adjustment are required, with a significant speed reduction of 10%-20% of the current speed. This allows sufficient time for the ship to adjust its course and correct its navigation, preventing further deviations that could lead to navigation accidents. The direction and magnitude of the speed adjustment should be combined with the ship's own power performance (such as maximum speed, minimum speed, and speed change response time), ship type (cargo ship, passenger ship, fishing ship), and cargo capacity (if it is a cargo ship) to develop a refined speed adjustment plan. The speed adjustment plan should be converted into a control signal and sent to the ship's power controller in real time.

[0064] After the ship's speed is dynamically adjusted and stabilized, the environmental status assessment phase begins. The core process involves real-time collection of environmental status data within the ship's navigation area. This data primarily includes four categories: meteorological data (wind force, wind direction, visibility, precipitation), waterway environmental data (waterway width, water depth, traffic density, waterway congestion), hydrological environmental data (water flow speed, water flow direction, tides), and special control environmental data (temporary waterway control, vessel traffic around service areas, berth availability at service areas). The data collection channels for each type of environmental data are clearly defined: meteorological data is acquired in real-time through the ship's onboard meteorological sensors and surrounding waterway meteorological monitoring stations (collected every 15 seconds); waterway environmental data is acquired collaboratively through an electronic chart system, radar detectors, and video signal capture unit 1; hydrological environmental data is acquired in real-time through waterway hydrological monitoring points; and special control environmental data is acquired in real-time through the ship dispatching system and the service area management platform. Fuzzy comprehensive evaluation technology is used to determine the weights of each type of environmental data and combine them with the environmental data to calculate the environmental status. The assessment score includes a focused evaluation of the environmental conditions within a 1000-meter radius of the service area, with particular attention paid to monitoring traffic density, vessel berthing order, and berth availability. This provides accurate environmental references for subsequent arrival time calculations. After the environmental condition assessment, the arrival time calculation phase begins. This phase obtains the vessel's current real-time latitude and longitude coordinates (output by the signal fusion trajectory prediction unit 2), the adjusted stable target speed (output by the speed dynamic adjustment unit), and the latitude and longitude coordinates of the service area. The straight-line distance between the vessel's current position and the service area is calculated. During the calculation, the remaining navigation distance is corrected based on the environmental condition assessment score and level. The theoretical arrival time is then calculated. Further corrections are made based on the environmental condition level and the surrounding environment of the service area. These corrections include environmental impact correction, berthing and waiting corrections around the service area, and channel congestion correction. After these corrections, the final estimated arrival time is obtained. Combined with the vessel's current real-time time, the estimated arrival time of the vessel at the service area is calculated.

[0065] The tracking and monitoring control unit 5 is used to acquire the temperature and navigation conditions in the navigation area during the expected period, analyze the consumption rate of the ship's hydraulic oil, and verify and optimize the ship's navigation speed based on the consumption rate to ensure the normal supply of the ship's hydraulic oil.

[0066] In one embodiment, the tracking and monitoring control unit 5 includes: The derivation relationship establishment module is used to acquire and classify temperature data and navigation conditions within the expected navigation period according to temperature gradient and traffic complexity, and establish a consumption rate derivation relationship based on temperature data and navigation conditions. The supply duration estimation module is used to obtain the consumption rate of ship hydraulic oil by using the consumption rate derivation relationship, and to estimate the supply duration of the ship under the consumption rate condition based on the total amount of ship hydraulic oil carried and the adjusted sailing speed. The sailing speed adaptation assessment module is used to determine the adaptation degree of the adjusted sailing speed based on the supply time, and to identify the deficiencies of the adjusted sailing speed based on the adaptation degree. It adopts segmented verification and linkage adjustment logic to optimize the sailing speed to ensure the supply quality of ship hydraulic oil.

[0067] In one embodiment, establishing the consumption rate derivation relationship based on temperature data and navigation conditions includes: using navigation speed as the independent variable and temperature data and navigation conditions as dependent variables, deriving the ship's idling speed during navigation and the speed influence coefficient corresponding to the ship's idling speed using a linear fitting method; adding the speed influence coefficient to a bidirectional search algorithm to analyze the average hydraulic oil consumption volume and average load of each dependent variable, and calculating the instantaneous consumption rate corresponding to each working condition in navigation conditions based on the average hydraulic oil consumption volume and average load; establishing a functional equation to describe the relationship between the instantaneous consumption rate and the change in navigation speed, and obtaining the consumption rate derivation result.

[0068] Among them, the signal fusion trajectory prediction unit 2, the trajectory compliance assessment unit 3, the arrival time prediction unit 4, and the tracking and monitoring control unit 5 constitute a closed-loop monitoring logic to complete the automatic prevention and control tracking and monitoring of the ship during navigation.

[0069] It should be explained that the tracking and monitoring control unit 5, as the core closed-loop control unit of the ship automatic prevention and control tracking and monitoring system, has the core objective of acquiring the temperature and navigation conditions of the navigation area during the ship's expected arrival time, analyzing the ship's hydraulic oil consumption rate, verifying and optimizing the ship's speed based on the consumption rate, ensuring the normal supply of hydraulic oil and eliminating the risk of shortage, and providing hydraulic system support for the ship's continuous and safe navigation. It also receives the ship's expected arrival time, expected navigation route, and navigation area from the arrival time prediction unit 4 in real time, and simultaneously connects with the ship's onboard temperature sensors and waterway meteorological monitoring stations to obtain the expected navigation time... (i.e., the entire period from the current moment to the expected arrival at the water service area) real-time and predicted temperature data within the navigation area, and simultaneously connect with electronic charts, radar detectors and video signal capture unit 1 to obtain navigation condition data for the navigation area within the expected period. Navigation conditions are mainly divided according to indicators such as navigation density, channel congestion, channel type, and avoidance frequency, specifically into three categories: smooth conditions (navigation density ≤ 1 vessel / km, no congestion, no need for frequent avoidance), general conditions (1 vessel / km < navigation density ≤ 3 vessels / km, slight congestion, occasional avoidance), and busy conditions (navigation density > 3 vessels / km, significant congestion, frequent avoidance).

[0070] The collected temperature data and navigation conditions were divided according to temperature gradient and navigation complexity. The temperature gradient was divided into five intervals of 5℃ each: below 0℃, 0-5℃, 5-10℃, 10-15℃, and above 15℃. Navigation complexity was divided according to the type of operation: low complexity for unobstructed conditions, medium complexity for general conditions, and high complexity for busy conditions. After the division, the temperature data, navigation conditions, and the previously adjusted ship speed were correlated. With ship speed as the independent variable and temperature data and navigation conditions as the dependent variables, a linear fitting method was used to derive the ship's idle speed during navigation using the least squares method. This idle speed is the engine speed when the ship maintains its basic navigation attitude and is not moving forward, expressed in r / min. The speed influence coefficient corresponding to the idle speed was also derived. This coefficient reflects the degree of influence of the ship's speed deviating from idle speed on hydraulic oil consumption; the larger the coefficient, the more significant the impact of speed change on the consumption rate. During the linear fitting process, invalid data with abnormal temperatures and sudden changes in navigation conditions were removed to ensure the accuracy of the derived results. The obtained speed influence coefficient is added to the bidirectional search algorithm. The core of the bidirectional search algorithm is to simultaneously search from the maximum and minimum speed values ​​towards the middle, analyzing the average hydraulic oil consumption volume (in L / h) and average load (in kW) corresponding to each dependent variable (different temperature gradients, different navigation conditions). The average consumption volume is calculated using real-time data collected by the ship's hydraulic oil sensors. The average load is calculated based on the ship's power output demand corresponding to the navigation conditions. The average load is 30%-40% of the ship's rated load under smooth conditions, 50%-60% under normal conditions, and 70%-80% under busy conditions. Based on the average hydraulic oil consumption volume and average load, the instantaneous consumption rate corresponding to each navigation condition is calculated using the formula v = (V × P) / (t × P0), where v represents the instantaneous consumption rate in L / min; V represents the average consumption volume; P represents the average load; t represents the data collection time; and P0 is the ship's rated load. After the calculation, a functional equation is established to describe the relationship between the instantaneous consumption rate and the change in sailing speed, i.e., v = k × v 航 +b×T+c×G, where v represents the instantaneous consumption rate, k represents the velocity influence coefficient, and v 航 Let b represent the sailing speed, T represent the temperature influence coefficient, c represent the navigation condition influence coefficient, and G represent the condition coefficient. For smooth navigation, G=1; for normal navigation, G=2; and for busy navigation, G=3. By substituting the measured data under different temperatures and conditions, the specific values ​​of b and c in the equation are solved, and the consumption rate derivation result is finally obtained, which is the complete consumption rate derivation relationship.

[0071] Based on the consumption rate derivation relationship, and substituting the adjusted target ship speed, the average temperature during the expected sailing period (the average of all temperature data within the expected period), and the average navigation conditions (the navigation conditions with the highest proportion during the expected period), the instantaneous hydraulic oil consumption rate under the ship's current sailing state is calculated. Simultaneously, the level sensor on the ship's hydraulic oil storage tank is used to obtain the total hydraulic oil carried by the ship (the preset total hydraulic oil carry is 200L, which can be updated in real time via the sensor; if there is consumption, it decreases synchronously). Combined with the adjusted ship speed, the hydraulic oil supply duration under this consumption rate condition is estimated. During the estimation process, the expected arrival time needs to be considered, and the supply duration is compared with... A preliminary comparison is made with the estimated arrival time. If the supply time is ≥ estimated arrival time + 10 minutes (with a 10-minute safety margin), it is preliminarily determined that the hydraulic oil supply is sufficient and there is no need to adjust the speed immediately. If the supply time is < estimated arrival time + 10 minutes, it is determined that there is a potential shortage of hydraulic oil and it needs to be marked as a key verification target. During the estimation process, the hydraulic oil consumption rate and remaining oil volume data need to be collected in real time. The supply time is re-estimated every 10 seconds, and the estimation results are dynamically updated. If there is a sudden change in temperature or a deterioration in navigation conditions (such as changing from normal conditions to busy conditions), the new temperature and operating condition data are immediately re-introduced to correct the instantaneous consumption rate and ensure the real-time and accurate estimation of the supply time.

[0072] Simultaneously, the suitability of the adjusted sailing speed is determined based on the supply time. The suitability judgment standard is divided into three levels: suitable, that is, the supply time ≥ the estimated arrival time + 10 minutes, the hydraulic oil supply is sufficient, and the speed does not need to be adjusted; basically suitable, that is, the estimated arrival time ≤ the supply time < the estimated arrival time + 10 minutes, the hydraulic oil supply is basically sufficient, but the speed needs to be slightly optimized; unsuitable, that is, the supply time < the estimated arrival time, the hydraulic oil cannot support the journey to the floating service area, and the speed needs to be significantly optimized. The inadequacy of the adjusted sailing speed is determined according to the suitability level. If it is the suitable level, it means that the current speed matches the hydraulic oil consumption rate, and the current speed is maintained unchanged while the consumption rate and supply time are continuously monitored. If it is the basically suitable or unsuitable level, it means that the hydraulic oil consumption rate corresponding to the current speed is too high, and the sailing speed needs to be optimized by using segmented verification and linkage adjustment logic.

[0073] The segmented verification and linkage adjustment logic is as follows: The system is divided into segments based on the temperature gradient and navigation conditions within the expected navigation period. Each segment corresponds to an independent verification unit. First, the consumption rate and supply duration of each segment are individually verified. The optimal speed required for that segment (i.e., the speed with the lowest consumption rate and guaranteed supply duration) is calculated. The optimal speed is calculated using a bidirectional search algorithm within a preset speed range (from the ship's minimum speed to the current target speed) to find the speed value that minimizes the consumption rate and meets the navigation requirements for that segment. The optimal speeds of each segment are then linked and integrated to ensure smooth speed adjustment. Simultaneously, the tracking compliance assessment unit 3 is linked to verify whether the optimized speed meets tracking compliance requirements (e.g., whether the optimized speed will lead to a decrease in tracking compliance). If it does, it is determined as the final optimized navigation speed. If it does not, the speed optimization range is readjusted to minimize the hydraulic oil consumption rate while ensuring tracking compliance, until the optimal navigation speed that meets both hydraulic oil supply requirements and tracking compliance requirements is found. After optimization, the new navigation speed is converted into a control signal and sent to the ship's power controller to synchronously update the ship's target speed.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A ship automatic prevention, control, tracking and monitoring system based on the fusion of AIS, radar and video signals, characterized in that, The system includes: The video signal capture unit is used to adjust the tracking state of the camera capture device according to the latitude, longitude and angle adjustment technology of the camera capture device, and to capture the video signal of the ship during the navigation process using the camera capture device after the tracking state adjustment. The signal fusion trajectory prediction unit interacts with the video signal capture unit to fuse ship digital signals, radar digital signals and video signals, and predict the ship's trajectory within the target time period based on the signal fusion results. The track compliance assessment unit is used to obtain the route to be navigated by the ship and calculate the offset distance between the ship's operating trajectory and the route to be navigated. Based on the offset distance, the track compliance of the ship during the navigation process is assessed. The arrival time prediction unit is used to adjust the ship's speed according to the compliance of the navigation track, and to assess the expected time when the ship will arrive at the water service area based on the adjusted ship speed and the environmental conditions in the navigation area. The tracking and monitoring control unit is used to acquire the temperature and navigation conditions in the navigation area during the expected period, analyze the consumption rate of the ship's hydraulic oil, and verify and optimize the ship's navigation speed based on the consumption rate to ensure the normal supply of the ship's hydraulic oil. The signal fusion trajectory prediction unit, the trajectory compliance assessment unit, the arrival time prediction unit, and the tracking and monitoring control unit constitute a closed-loop monitoring logic to complete the automatic prevention, control, tracking, and monitoring of ships during navigation.

2. The ship automatic prevention and control tracking and monitoring system based on AIS, radar, and video signal fusion as described in claim 1, characterized in that, The signal fusion trajectory prediction unit includes: The digital signal acquisition module is used to acquire the ship's digital signals and radar digital signals during navigation using the Automatic Identification System (AIS) and radar detector, respectively. The signal fusion output module is used to calculate the distance between the ship's digital signal, radar digital signal and video signal, determine the corresponding trust function, and perform signal fusion processing by combining fuzzy clustering technology to output fused signal data. The navigation trajectory prediction module is used to combine fused signal data with ship route maps to output a set of candidate trajectory endpoints, and generate a heat map based on the set of candidate trajectory endpoints to predict the ship's operating trajectory within the target time period.

3. The ship automatic prevention and control tracking monitoring system based on AIS, radar, and video signal fusion as described in claim 2, is characterized in that, The process involves calculating the distances between the ship's digital signals, radar digital signals, and video signals, determining the corresponding trust functions, and performing signal fusion processing using fuzzy clustering techniques. The output fused signal data includes: Outlier removal and normalization are performed on ship digital signals, radar digital signals and video signals, and the distance between any two types of signals is calculated after processing to form a distance matrix; The element similarity between ship digital signals, radar digital signals and video signals is calculated based on the distance matrix. After normalizing the element similarity, the trust function between any two types of signals is calculated. Remove signal data from ship digital signals, radar digital signals, and video signals whose trust function is lower than the trust threshold, and calculate the mean of the remaining signal data to output outlier data; Principal component analysis is used to process heterogeneous data, and a similarity matrix between heterogeneous data is calculated based on the processing results, so as to capture the relationship between heterogeneous data using the similarity matrix; Based on the similarity matrix, the clustering radius is set, and the optimal solution is iteratively screened to determine the optimal solution. The optimal solution is then used to calculate the weight vector to generate the membership matrix, which is used to fuse heterogeneous data and output fused signal data.

4. The ship automatic prevention and control tracking and monitoring system based on AIS, radar, and video signal fusion as described in claim 3, is characterized in that, The process of using fused signal data combined with a ship route map to output a set of candidate trajectory endpoints, and generating a heat map based on the candidate trajectory endpoint set to predict the ship's trajectory within the target time period includes: Based on fused signal data and combined with ship route map information, graph neural network technology is used to jointly encode the ship's trajectory information and ship route map information to obtain scene coding vector; The set of candidate trajectory points of the ship in the navigation area is obtained based on the scene encoding vector, and the probability value of each candidate trajectory point in the set within the target time period is analyzed by combining the attention mechanism. A heat map is generated based on the probability values ​​of each candidate trajectory point, presenting the probability of a ship appearing in the navigation area within the target time period. Based on the occurrence probability and the trajectory completion processing method, the ship's operating trajectory within the target time period is predicted and the ship's operating trajectory is determined.

5. The ship automatic prevention and control tracking monitoring system based on AIS, radar, and video signal fusion as described in claim 4, characterized in that, The method involves using fused signal data combined with ship route map information, and employing graph neural network technology to jointly encode the ship's trajectory information and ship route map information, resulting in a scene encoding vector including: The fused signal data is split into trajectory vectors connected end to end in chronological order, and the ship route map information is defined as map vectors to describe the geographical features of ship navigation. By using trajectory vectors and map vectors as graph nodes and adjacent vectors as edges, node information is transferred and local features are extracted to obtain the encoded features of the local map of ship trajectory and the local map. The encoded features are used as input to perform global information fusion through a graph neural network to construct a global scene graph containing the ship's historical trajectory and navigation environment, resulting in a two-dimensional scene encoded feature matrix. The fused signal data is embedded into standardized coordinates to obtain an embedding vector. The embedding vector is used to represent the coordinate encoding information of the ship at different times. The embedding vector is then input into the encoder sequentially according to each time to extract the temporal features of the ship's motion state. The temporal features output by the encoder are input into a fully connected layer containing an activation function to obtain the encoded vector of the ship's motion state.

6. The ship automatic prevention, control, tracking and monitoring system based on AIS, radar and video signal fusion as described in claim 5, characterized in that, The process of obtaining a set of candidate trajectory points for the ship within the navigation area based on scene encoding vectors, and analyzing the probability value of each candidate trajectory point in the target time period using an attention mechanism, includes: Based on the ship route map information, the navigation area is determined. Within the navigation area, the target prediction time period and the ship's navigation speed operation scenario coding map are combined to determine the ship's trajectory points within the navigation area. Candidate trajectory points are integrated based on a dense sampling method, and candidate points that are more than a threshold distance and conflict with obstacles in the navigation area are removed to obtain a set of candidate trajectory points. The two-dimensional scene coding feature matrix and coding vector are fused. The coding vectors related to ships in the two-dimensional scene coding feature matrix are superimposed, and the row vectors that are not related are filled with zero vectors to obtain the fused matrix. Based on the fusion matrix and the two-dimensional scene coding feature matrix, a comprehensive coding matrix containing information on ship navigation scene and operation status is determined. The comprehensive coding matrix and the candidate trajectory point set are then mapped to a query matrix, a key matrix, and a value matrix using a projection matrix. Based on the query matrix, key matrix, and value matrix, the attention distribution of the association weights between candidate trajectory points and encoding vectors is calculated using the scaling point attention mechanism. Based on the attention distribution, the trajectory driving candidate points after fusing attention features are obtained and input into the fully connected layer to output the probability value of the trajectory driving candidate points in the target time period.

7. The ship automatic prevention and control tracking monitoring system based on AIS, radar, and video signal fusion as described in claim 1, characterized in that, The flight track compliance assessment unit includes: The route division module is used to obtain the route to be navigated by the ship, divide the route to be navigated by the ship into three stages according to the route status, and determine the anchoring feature points of each segment of the route to be navigated by the ship. The three stages include a mandatory segment, an adjustment segment, and a warning segment. The offset distance definition module is used to synchronize and align the ship's trajectory with the segmented results of the ship's route according to time nodes, and define the offset distance between the ship's trajectory and the route to be navigated based on anchor feature points and segment comparison logic. The navigation compliance judgment module is used to assign weights to the offset distance of each segment of the ship's route according to the navigation environment. Based on the offset distance and the corresponding weights, the module determines the compliance of the ship's track during navigation by logical superposition.

8. The ship automatic prevention and control tracking and monitoring system based on AIS, radar, and video signal fusion as described in claim 7, characterized in that, In the segmentation comparison logic, when the segmentation result of the ship's navigation route is a forced segment, the straight-line offset distance between the ship's trajectory and the baseline is defined based on the anchoring feature point. In the segmentation comparison logic, when the segmentation result of the ship's route to be navigated is an adjustment segment, the preset offset threshold range is used as a reference to determine whether the ship's trajectory is within the allowable offset range of the preset offset threshold range. In the segmented comparison logic, when the segmented result of the ship's route to be navigated is a warning segment, the ship's trajectory is defined as the offset distance and direction from the warning critical point, using the warning critical point as the standard.

9. The ship automatic prevention and control tracking monitoring system based on AIS, radar, and video signal fusion as described in claim 1, characterized in that, The tracking and monitoring control unit includes: The derivation relationship establishment module is used to acquire and classify temperature data and navigation conditions within the expected navigation period according to temperature gradient and traffic complexity, and establish a consumption rate derivation relationship based on temperature data and navigation conditions. The supply duration estimation module is used to obtain the consumption rate of ship hydraulic oil by using the consumption rate derivation relationship, and to estimate the supply duration of the ship under the consumption rate condition based on the total amount of ship hydraulic oil carried and the adjusted sailing speed. The sailing speed adaptation assessment module is used to determine the adaptation degree of the adjusted sailing speed based on the supply time, and to identify the deficiencies of the adjusted sailing speed based on the adaptation degree. It adopts segmented verification and linkage adjustment logic to optimize the sailing speed to ensure the supply quality of ship hydraulic oil.

10. The ship automatic prevention, control, tracking and monitoring system based on AIS, radar and video signal fusion as described in claim 9, characterized in that, The relationship between consumption rate and temperature data and air traffic conditions is established as follows: Using sailing speed as the independent variable and temperature data and navigation conditions as the dependent variables, a linear fitting method was used to derive the ship's idling speed during navigation and the speed influence coefficient corresponding to the ship's idling speed. The speed influence coefficient is added to the bidirectional search algorithm to analyze the average hydraulic oil consumption volume and average load of each dependent variable, and the instantaneous consumption rate corresponding to each working condition in the navigation situation is calculated based on the average hydraulic oil consumption volume and average load. A functional equation was established to describe the relationship between the instantaneous consumption rate and the change in sailing speed, and the derivation result of the consumption rate was obtained.

Images