Radar and AIS track association method, device and equipment based on spatiotemporal attention

Through the method based on space-time attention, multi-layer perceptron and Transformer encoder extracting track features, combined with cross-attention mechanism, the problem of poor accuracy of traditional correlation methods in complex environments is solved, and multiple complex waters are achieved The intelligent association fusion of ship information improves association accuracy and adaptability.

CN118818485BActive Publication Date: 2024-11-19WUHAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411302814.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-19
Estimated Expiration
2044-09-19

AI Technical Summary

Technical Problem

In complex maritime environments, it is difficult to achieve accurate matching between traditional radar and AIS track correlation methods, resulting in poor correlation accuracy. Especially in dense waters and ship situations, error correlation is prone to occur, and multiple empirical parameters are required to be set. It affects the accuracy of the association and cannot fully adapt to the changing marine environment.

Method used

Using a method based on space-time attention, the local spatial characteristics of the track data are extracted through multi-layer perception machines, combined with position encoding and Transformer encoder to extract global spatio-time characteristics, and used the cross attention mechanism to calculate the correlation probability matrix to realize radar and AIS tracks. exact match.

Benefits of technology

It improves correlation accuracy and scenario adaptability, can intelligently associate multiple ship information in complex waters, improves the intelligence level of ship monitoring and channel management, and has strong robustness and generalization capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118818485B_ABST
    Figure CN118818485B_ABST
Patent Text Reader

Abstract

A radar and AIS track association method, device and equipment based on spatiotemporal attention, the method includes: obtaining radar track data and AIS track data; using a multi-layer perceptron to enhance the dimension of the track data and extract the local spatial features of the track data; encoding the track data by position to obtain the time series features of the track data; splicing the local spatial features of the track data with the time series features of the track data, and inputting the spliced ​​features into the Transformer encoder to encode the global spatiotemporal features of the track data to obtain the spatiotemporal feature encoding vector of the track data; using a cross-attention mechanism to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain an association probability matrix; and associating and matching the radar track with the AIS track based on the association probability matrix. The present invention can realize the intelligent association and fusion of multiple ship information in complex waters, with stronger scene adaptability and higher association accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source information fusion, and in particular to a radar and AIS track association method, device and equipment based on spatiotemporal attention. Background Art

[0002] In densely populated scenarios such as ports, busy waters and near-shore fishing areas, the risk of ship collisions continues to rise. Using advanced navigation information perception technology to track and monitor ship positions in real time can effectively identify potential risks and issue warnings to reduce collisions. Single sensors have problems with missed detection and false alarms. Radar detection is affected by ship size, sea conditions and weather, making it difficult to accurately track and identify multiple targets. The Automatic Identification System (AIS) provides accurate ship location information, but can only identify ships equipped with AIS equipment, and the information is updated infrequently. Integrate effective radar and AIS information to achieve comprehensive real-time perception in complex environments, improve positioning accuracy, and ensure maritime navigation safety.

[0003] The traditional association method builds a motion model for the AIS target, iterates and updates it to keep time synchronization with the radar track, and then associates it. However, it is difficult to establish an accurate motion model in a complex marine environment, resulting in poor association accuracy. Another method is to first calculate the similarity between the radar track and the AIS track, and then perform data association based on the calculated track similarity to achieve the purpose of fusion. However, in dense waters and when ships encounter complex situations, the above method will result in incorrect association. In addition, the above association method requires the setting of many empirical parameters. The appropriateness of the empirical parameters directly affects the association accuracy, and the same set of empirical parameters cannot fully adapt to the complex and changeable marine environment. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned defects and problems existing in the prior art, and to provide a radar and AIS track association method, device and equipment based on spatiotemporal attention, so as to realize the intelligent association and fusion of multiple ship information in complex waters, with stronger scene adaptability and higher association accuracy.

[0005] To achieve the above objectives, the technical solution of the present invention is:

[0006] A radar and AIS track correlation method based on spatiotemporal attention, comprising:

[0007] Acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship's navigation detected by the radar within a time window, and the AIS track data is the track data of the ship's navigation within the radar coverage area recorded by the AIS system;

[0008] Use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data;

[0009] Position-encode the track data to obtain the time series characteristics of the track data;

[0010] The local spatial features of the track data are concatenated with the temporal features of the track data, and the concatenated features are input into the Transformer encoder to encode the global spatiotemporal features of the track data, thereby obtaining the spatiotemporal feature encoding vector of the track data;

[0011] The cross attention mechanism is used to calculate the spatiotemporal feature encoding vectors of radar track data and AIS track data to obtain the association probability matrix;

[0012] The radar tracks are correlated and matched with the AIS tracks based on the correlation probability matrix.

[0013] The radar track data for:

[0014] ;

[0015] ; ;

[0016] In the formula, For the Target tracks detected by radar; , , and They are the first The longitude, latitude, ground speed and ground heading information of each track point; is the total number of target tracks detected by the radar within a time window;

[0017] The AIS track data for:

[0018] ;

[0019] ; ;

[0020] In the formula, The radar coverage area AIS target tracks; , , and The first The longitude, latitude, ground speed and ground heading information of each track point; It is the total number of AIS target tracks within the radar coverage area within a time window.

[0021] Pre-process radar track data and AIS track data, including:

[0022] Normalize the radar track data and AIS track data respectively;

[0023] Padding processing is performed on the radar track data and the AIS track data to make the lengths of the radar track data and the AIS track data consistent.

[0024] Using an improved multi-layer perceptron to increase the dimension of the track data, the improved multi-layer perceptron includes three hidden layers, and a nonlinear activation function and a Dropout layer are added after each hidden layer;

[0025] The track data is position encoded using the following formula:

[0026] ;

[0027] ;

[0028] In the formula, Indicates positional encoding; Indicates the position of a track point in a section of track data; Represents the index of each element in the vector; Represents the dimension of the position encoding vector.

[0029] The cross attention mechanism is used to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain an association probability matrix, including:

[0030] The attention scores of the spatiotemporal feature coding vectors of the radar track data and the spatiotemporal feature coding vectors of the AIS track data are calculated to obtain the similarity score matrix between the radar track data and the AIS track data. , for the similarity score matrix The rows and columns of are processed by Softmax respectively to obtain the normalized probability matrix and , for the probability matrix and Take the average value to get the association probability matrix ;

[0031] The calculation formula of attention score is:

[0032] ;

[0033] In the formula, is the calculation function of the attention score; The spatiotemporal feature encoding vector representing radar track data ; Represents the spatiotemporal feature encoding vector of AIS track data ; is the length of the track data; is a matrix of all 1s.

[0034] Constructing the representation probability matrix and With the real correlation matrix The loss function is the sum of the cross entropy of , and the expression of the loss function is:

[0035] ;

[0036] ; ;

[0037] ;

[0038] In the formula, is the standard cross entropy loss with weight factors; Represents the predicted probability of the true category of the sample; is a hyperparameter for adjusting the weight factor.

[0039] The method of associating and matching the radar track with the AIS track based on the association probability matrix includes:

[0040] The association probability matrix The value in each row is the largest and greater than the associated confidence threshold Assign a value of 1, otherwise 0, and get an association matrix ; Correlation matrix Include elements , Indicates the input The radar track AIS tracks are a pair of related tracks from the same target.

[0041] A radar and AIS track association device based on spatiotemporal attention, which is applied to the above-mentioned method, and comprises:

[0042] A track data acquisition module, used to acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship sailing detected by the radar within a time window, and the AIS track data is the track data of the ship sailing within the radar coverage area recorded by the AIS system;

[0043] The association probability matrix acquisition module is used to use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data; perform position encoding on the track data to obtain the time series features of the track data; concatenate the local spatial features of the track data with the time series features of the track data, and input the concatenated features into the Transformer encoder to encode the global spatiotemporal features of the track data to obtain the spatiotemporal feature encoding vector of the track data; use the cross-attention mechanism to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain the association probability matrix;

[0044] The track correlation matching module is used to correlate and match the radar track with the AIS track based on the correlation probability matrix.

[0045] A radar and AIS track correlation device based on spatiotemporal attention, comprising a memory and a processor;

[0046] The memory is used to store computer program code and transmit the computer program code to the processor;

[0047] The processor is used to execute the method according to the instructions in the computer program code.

[0048] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] In the present invention, a radar and AIS track association method, device and equipment based on spatiotemporal attention are provided. The method uses a multi-layer perceptron to perform high-dimensional characterization on track data to obtain local spatial features of the track, uses position coding to add temporal features to the track, and splices them with the local spatial features of the track, and uses a Transformer encoder to extract and encode the global spatiotemporal features of the track data; in the association stage, a cross-attention mechanism is used to calculate the association probability matrix, thereby achieving accurate matching of radar targets and AIS targets. The present invention can realize intelligent association and fusion of multiple ship information in complex waters, with stronger scene adaptability and higher association accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The present invention is a flow chart of a radar and AIS track association method based on spatiotemporal attention.

[0052] Figure 2 It is a comparison diagram between the real association matrix and the predicted association matrix in an embodiment of the present invention.

[0053] Figure 3 This is a diagram showing the actual fusion effect of radar and AIS tracks based on spatiotemporal attention in an embodiment of the present invention.

[0054] Figure 4 The present invention is a structural block diagram of a radar and AIS track association device based on spatiotemporal attention.

[0055] Figure 5 The present invention is a structural block diagram of a radar and AIS track association device based on spatiotemporal attention. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0057] See also Figure 1 , a radar and AIS track association method based on spatiotemporal attention, comprising:

[0058] S1. Acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship sailing detected by the radar within a time window, and the AIS track data is the track data of the ship sailing within the radar coverage area recorded by the AIS system;

[0059] S2, use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data;

[0060] Position-encode the track data to obtain the time series characteristics of the track data;

[0061] The local spatial features of the track data are concatenated with the temporal features of the track data, and the concatenated features are input into the Transformer encoder to encode the global spatiotemporal features of the track data, thereby obtaining the spatiotemporal feature encoding vector of the track data;

[0062] The cross attention mechanism is used to calculate the spatiotemporal feature encoding vectors of radar track data and AIS track data to obtain the association probability matrix;

[0063] S3. Correlate and match the radar track with the AIS track based on the correlation probability matrix.

[0064] Compared with traditional association methods, the present invention achieves more accurate target association through the spatiotemporal attention mechanism, and can automatically learn the dynamic characteristics and measurement errors of the target in a complex marine environment, thereby realizing the intelligent association and fusion of multiple ship information in complex waters, making the data association scene more adaptable, and the association accuracy greatly improved, effectively improving the intelligent level of ship monitoring and waterway management. In addition, the method also has strong robustness and generalization ability, and can be widely used in multi-source asynchronous track data association tasks in different scenarios.

[0065] Furthermore, radar track data for:

[0066] ;

[0067] ; ;

[0068] In the formula, For the Target tracks detected by radar; , , and They are respectively the longitude, latitude, speed over the ground and course over the ground of the first track point of the ship target detected by the radar; , , and They are the first The longitude, latitude, ground speed and ground heading information of each track point; It is the total number of target tracks detected by the radar within a time window.

[0069] Furthermore, AIS track data for:

[0070] ;

[0071] ; ;

[0072] In the formula, The radar coverage area AIS target tracks; , , and The longitude, latitude, speed over the ground and course over the ground of the first track point of the ship target recorded by the AIS system; , , and The first The longitude, latitude, ground speed and ground heading information of each track point; It is the total number of AIS target tracks within the radar coverage area within a time window.

[0073] Furthermore, the radar track data and AIS track data are preprocessed so that the network model can effectively utilize the multi-feature information of the track data, including:

[0074] Normalize the radar track data and AIS track data respectively;

[0075] Padding processing is performed on the radar track data and the AIS track data to make the lengths of the radar track data and the AIS track data consistent.

[0076] Due to the different sampling frequencies of radar track data and AIS track data, the number of radar track points and AIS track points of the same target in a time window is different. Therefore, it is necessary to fill the radar track data and AIS track data to unify the length of the radar track data and AIS track data to , which facilitates the subsequent construction and training of neural networks.

[0077] Furthermore, an improved multi-layer perceptron is used to increase the dimension of the track data. In the present application, the dimension of the track data is greater than 4, so as to perform high-order characterization on the track data and extract the local spatial features of the track data. The improved multi-layer perceptron includes three hidden layers, and a nonlinear activation function and a Dropout layer are added after each hidden layer. The nonlinear activation function can enhance the network model's ability to extract local high-dimensional nonlinear spatial features of features such as position, speed and heading in the track data. The Dropout layer randomly discards 20% of neurons to prevent overfitting and simplify calculations.

[0078] Furthermore, the track data is position-encoded using the following formula:

[0079] ;

[0080] ;

[0081] In the formula, Indicates positional encoding; Represents the position of a track point in a segment of track data. The position code of each track point is a vector. Represents the index of each element in the vector; Represents the dimension of the position encoding vector.

[0082] The above formula can be used to represent the absolute position relationship of track points in a section of track data as a position encoding vector, so that the network model can capture the temporal characteristics between track points in a section of track data. The local spatial features of the track data are concatenated with the temporal features after position encoding, and the features are input into the Transformer encoder to encode the global spatiotemporal features of the track, thereby obtaining the spatiotemporal feature encoding vector of the radar track data. , , and the spatiotemporal feature encoding vector of AIS track data , .

[0083] Furthermore, the cross-attention mechanism is used to calculate the spatiotemporal feature encoding vectors of the radar track data and the spatiotemporal feature encoding vectors of the AIS track data to obtain the association probability matrix, including:

[0084] The attention scores of the spatiotemporal feature coding vectors of the radar track data and the spatiotemporal feature coding vectors of the AIS track data are calculated to obtain the similarity score matrix between the radar track data and the AIS track data. , , for the similarity score matrix The rows and columns of are processed by Softmax respectively to obtain the normalized probability matrix and , for the probability matrix and Take the average value to get the association probability matrix ;

[0085] The calculation formula of attention score is:

[0086] ;

[0087] In the formula, is the calculation function of the attention score; The spatiotemporal feature encoding vector representing radar track data ; Represents the spatiotemporal feature encoding vector of AIS track data ; is the length of the track data; is a matrix of all 1s, ; Indicates the transpose operation of the matrix. The purpose is to reduce The sensitivity of the length of the vector is reduced, so that no matter what the length of the vector is, the variance of the dot product remains 1 without considering the length of the vector, which facilitates the optimization of the network model and improves the stability of the network model during training.

[0088] By constructing the all-1 matrix as vector, and the spatiotemporal feature encoding vector of radar track data ( vector) and the spatiotemporal feature encoding vector of AIS track data ( This model design effectively improves the accuracy and efficiency of data association.

[0089] The present invention performs similarity score matrix The rows and columns of are Softmax normalized and then the average value is taken. This design makes the maximum element of each row in the association probability matrix also the largest element in the column where the element is located, avoiding the one-to-many association problem. At the same time, it can also avoid the wrong association of radar targets due to Softmax normalization when there is no corresponding AIS target.

[0090] In order to solve the technical problem of associating radar tracks and AIS tracks, the present invention combines the attention mechanism on the deep learning framework to construct a deep learning network model, that is, the network model described above. The network model uses a multi-layer perceptron to perform high-dimensional representation of the track data to obtain the local spatial features of the track, uses position coding to add time series features to the track, and splices them with the local spatial features of the track, and uses the Transformer encoder to extract and encode the global spatiotemporal features of the track data. The historical track data is input into the network model to obtain the predicted association probability matrix. In the association stage, the cross-attention mechanism is used to calculate the association probability matrix, thereby achieving accurate matching of radar targets and AIS targets.

[0091] Furthermore, construct the probability matrix and With the real correlation matrix The loss function is the sum of the cross entropy of , and the expression of the loss function is:

[0092] ;

[0093] ; ;

[0094] ;

[0095] In the formula, is the standard cross entropy loss with weight factors; represents the predicted probability of the true category of the sample, , is the predicted value, Indicates the true category of the sample; is a hyperparameter for adjusting the weight factor.

[0096] The differentiable loss function designed in the present invention enables the network model weights to be back-propagated, thereby obtaining a trained network model, which can ensure that there is at most one maximum value in each row and column of the association probability matrix to the greatest extent, thus avoiding the situation in the track association process where a radar track has no corresponding AIS track but is mistakenly associated with an unrelated AIS track, or a radar track is associated with multiple AIS tracks. Target association is regarded as a binary classification problem, and a weighted factor is designed. The loss function weakens the learning of easy samples and strengthens the learning of difficult samples, improves the classification imbalance problem in the association matrix samples, and improves the classification ability of the network model. The introduced weight factor adjusts the contribution of the sample to the loss according to the difference between the predicted result and the true label. When the difference is smaller, the prediction is more accurate, and the loss weight is reduced. Conversely, the loss weight is increased. Through back propagation and multiple rounds of training, the network model can achieve efficient and accurate target association matching.

[0097] Furthermore, the radar track and the AIS track are correlated and matched based on the correlation probability matrix, including:

[0098] The association probability matrix The value in each row is the largest and greater than the associated confidence threshold Assign a value of 1, otherwise 0, and get an association matrix , ; Correlation matrix is a 0-1 matrix consisting of the elements , Indicates the input The radar track AIS tracks are a pair of related tracks from the same target.

[0099] In view of the problems that the dynamic characteristics of the target in the traditional method are difficult to capture, the measurement error is large, the target motion model is difficult to establish, the time synchronization process is complex, and the correlation accuracy is low. At the same time, the two-stage correlation method based on track similarity calculation and track matching requires the setting of many empirical parameters, whose values ​​directly affect the correlation accuracy. The same set of empirical parameters cannot fully adapt to the complex and changeable marine environment, and the correlation accuracy is low. The present invention provides a radar and AIS track correlation method based on spatiotemporal attention to solve the problems of low robustness and generalization, difficult time synchronization, and low correlation accuracy in the existing technology under complex and changeable scenes. The present invention learns the spatiotemporal characteristics of the track data through the constructed spatiotemporal feature fusion coding network training, uses the attention mechanism to focus on the spatiotemporal characteristics, measures the similarity between tracks, and realizes the intelligent correlation and fusion of multiple ship information in complex waters, with stronger scene adaptability and higher correlation accuracy.

[0100] Figure 2 One part is the correlation matrix of radar tracks and AIS tracks obtained after manually calibrating the real correlation track pairs within a time window. There are 45 radar tracks and 43 AIS tracks. The black box in the figure is Represents the current time window. The radar track AIS tracks are a pair of related tracks from the same target; Figure 2 The other part is the correlation matrix predicted by the model proposed by the present invention The prediction accuracy in the current time window is 100%, and the prediction accuracy of the model in the entire test set is 96.89%.

[0101] Figure 3 This is the actual fusion effect diagram of the radar and AIS track fusion method based on spatiotemporal attention. In the figure, you can see that the ship representing the fused radar and AIS information includes the Maritime Mobile Service Identity (MMSI), ship name, call sign, radar track ID number and other information. The fused information can allow the relevant maritime management departments to obtain the static information of the target ship in the supervision area. At the same time, the radar can make up for the slow update of the target ship's AIS position, ensuring that the management department can track and supervise the department in real time.

[0102] The present invention does not require the establishment of a complex motion model or complex time synchronization processing, and associates multi-source asynchronous tracks through a trained network model. The present invention proposes an end-to-end association method. Compared with a two-stage association method that measures the similarity of tracks and then sets empirical parameters for data association, the method proposed by the present invention is simpler in practical applications and has higher association accuracy. In addition, the proposed end-to-end association method is not only suitable for the association between radar and AIS tracks, but also has good scalability and adaptability, and can be applied to other multi-source asynchronous track association scenarios, providing new ideas and methods for research and application in related fields.

[0103] See also Figure 4 , a radar and AIS track association device based on spatiotemporal attention, the device is applied to the above-mentioned radar and AIS track association method based on spatiotemporal attention, the device comprises:

[0104] A track data acquisition module, used to acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship sailing detected by the radar within a time window, and the AIS track data is the track data of the ship sailing within the radar coverage area recorded by the AIS system;

[0105] The association probability matrix acquisition module is used to use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data; perform position encoding on the track data to obtain the time series features of the track data; concatenate the local spatial features of the track data with the time series features of the track data, and input the concatenated features into the Transformer encoder to encode the global spatiotemporal features of the track data to obtain the spatiotemporal feature encoding vector of the track data; use the cross-attention mechanism to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain the association probability matrix;

[0106] The track correlation matching module is used to correlate and match the radar track with the AIS track based on the correlation probability matrix.

[0107] See also Figure 5 , a radar and AIS track correlation device based on spatiotemporal attention, including a memory and a processor;

[0108] The memory is used to store computer program code and transmit the computer program code to the processor;

[0109] The processor is used to execute the above-mentioned radar and AIS track association method based on spatiotemporal attention according to the instructions in the computer program code.

[0110] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-mentioned radar and AIS track association method based on spatiotemporal attention.

[0111] Generally speaking, the computer instructions for implementing the method of the present invention may be carried in any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media may include any computer-readable media except for the signal itself that is temporarily propagating.

[0112] Computer readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination thereof. More specific examples (non-exhaustive list) of computer readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EKROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0113] Computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer or to an external computer (for example, using an Internet service provider to connect via the Internet) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0114] The above-mentioned device and non-temporary computer-readable storage medium can refer to the specific description of a radar and AIS track association method based on spatiotemporal attention and its beneficial effects, which will not be repeated here.

[0115] Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A radar and AIS track association method based on spatiotemporal attention, characterized in that: include: Acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship's navigation detected by the radar within a time window, and the AIS track data is the track data of the ship's navigation within the radar coverage area recorded by the AIS system; Use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data; Position-encode the track data to obtain the time series characteristics of the track data; The local spatial features of the track data are concatenated with the temporal features of the track data, and the concatenated features are input into the Transformer encoder to encode the global spatiotemporal features of the track data, thereby obtaining the spatiotemporal feature encoding vector of the track data; The cross attention mechanism is used to calculate the spatiotemporal feature encoding vectors of radar track data and AIS track data to obtain the association probability matrix; The radar tracks are correlated and matched with the AIS tracks based on the correlation probability matrix.

2. A radar and AIS track association method based on spatiotemporal attention according to claim 1, characterized in that: The radar track data for: ; ; ; In the formula, For the Target tracks detected by radar; , , and They are the first The longitude, latitude, ground speed and ground heading information of each track point; is the total number of target tracks detected by the radar within a time window; The AIS track data for: ; ; ; In the formula, The radar coverage area AIS target tracks; , , and The first The longitude, latitude, ground speed and ground heading information of each track point; It is the total number of AIS target tracks within the radar coverage area within a time window.

3. A radar and AIS track association method based on spatiotemporal attention according to claim 1, characterized in that: Pre-process radar track data and AIS track data, including: Normalize the radar track data and AIS track data respectively; Padding processing is performed on the radar track data and the AIS track data to make the lengths of the radar track data and the AIS track data consistent.

4. The radar and AIS track association method based on spatiotemporal attention according to claim 1 is characterized in that: Using an improved multi-layer perceptron to increase the dimension of the track data, the improved multi-layer perceptron includes three hidden layers, and a nonlinear activation function and a Dropout layer are added after each hidden layer; The track data is position encoded using the following formula: ; ; In the formula, Indicates positional encoding; Indicates the position of a track point in a section of track data; Represents the index of each element in the vector; Represents the dimension of the position encoding vector.

5. The radar and AIS track association method based on spatiotemporal attention according to claim 1 is characterized in that: The cross attention mechanism is used to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain an association probability matrix, including: The attention scores of the spatiotemporal feature coding vectors of the radar track data and the spatiotemporal feature coding vectors of the AIS track data are calculated to obtain the similarity score matrix between the radar track data and the AIS track data. , for the similarity score matrix The rows and columns of are processed by Softmax respectively to obtain the normalized probability matrix and , for the probability matrix and Take the average value to get the association probability matrix ; The calculation formula of attention score is: ; In the formula, is the calculation function of the attention score; The spatiotemporal feature encoding vector representing radar track data ; Represents the spatiotemporal feature encoding vector of AIS track data ; is the length of the track data; is a matrix of all 1s.

6. A radar and AIS track association method based on spatiotemporal attention according to claim 5, characterized in that: Constructing the representation probability matrix and With the real correlation matrix The loss function is the sum of the cross entropy of , and the expression of the loss function is: ; ; ; ; In the formula, is the standard cross entropy loss with weight factors; Represents the predicted probability of the true category of the sample; is a hyperparameter for adjusting the weight factor.

7. A radar and AIS track association method based on spatiotemporal attention according to claim 5, characterized in that: The method of associating and matching the radar track with the AIS track based on the association probability matrix includes: The association probability matrix The value in each row is the largest and greater than the associated confidence threshold Assign a value of 1, otherwise 0, and get an association matrix ; Correlation matrix Include elements , Indicates the input The radar track AIS tracks are a pair of related tracks from the same target.

8. A radar and AIS track association device based on spatiotemporal attention, characterized in that: The device is applied to the method described in any one of claims 1 to 7, and the device comprises: A track data acquisition module, used to acquire radar track data and AIS track data, wherein the radar track data is the track data of the ship sailing detected by the radar within a time window, and the AIS track data is the track data of the ship sailing within the radar coverage area recorded by the AIS system; The association probability matrix acquisition module is used to use a multi-layer perceptron to increase the dimension of the track data and extract the local spatial features of the track data; perform position encoding on the track data to obtain the time series features of the track data; concatenate the local spatial features of the track data with the time series features of the track data, and input the concatenated features into the Transformer encoder to encode the global spatiotemporal features of the track data to obtain the spatiotemporal feature encoding vector of the track data; use the cross-attention mechanism to calculate the spatiotemporal feature encoding vector of the radar track data and the spatiotemporal feature encoding vector of the AIS track data to obtain the association probability matrix; The track correlation matching module is used to correlate and match the radar track with the AIS track based on the correlation probability matrix.

9. A radar and AIS track association device based on spatiotemporal attention, characterized in that: including memory and processor; The memory is used to store computer program code and transmit the computer program code to the processor; The processor is configured to execute the method according to any one of claims 1 to 7 according to instructions in the computer program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.