Lightweight radar positioning and trajectory prediction method and device for maritime ship movement

Abstract

The invention discloses a lightweight radar positioning and trajectory prediction method and device for marine ship movement, and belongs to the field of radar positioning. The method comprises the following steps: preprocessing sample time domain echo data according to a conventional beam forming method to obtain angle sample data, and generating a label of the angle sample data based on a known sample incident angle; angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output; taking the angle sample data as input, taking the label as output, and training the YOLO network by using a plurality of second samples; based on the time domain echo data received by the radar and a pre-trained YOLO network, determining the distance and the incident angle between the radar and the point target; and determining a three-dimensional coordinate of the point target by using the determined distance and incident angle, and completing radar positioning of marine ship movement. The YOLO network provided by the invention has a lightweight structure, can be applied to small equipment, and can also ensure the positioning accuracy and real-time performance.

Description

Technical Field

[0001] The present invention relates to the technical field of radar positioning, and particularly relates to a lightweight radar positioning and trajectory prediction method and device for the movement of maritime ships. Background Art

[0002] In addition to sailing along the driven track, a maritime ship is also affected by the undulation of the time-varying sea surface, resulting in attitude angle changes in six degrees of freedom. The attitude angle changes under high sea conditions are particularly complex. Whether using high-performance electromagnetic calculation methods to simulate the radar scattering characteristics of maritime ships and then performing radar imaging under the determined track and attitude, or using test methods to detect the radar echo of maritime ships and using AK technology for radar imaging, it is necessary to know in advance the attitude changes of the ship target. The real-time and uncertainty of the target attitude changes make it very difficult to analyze the target characteristics of maritime ships.

[0003] In related technologies, a neural network method is used to achieve radar detection of the movement of maritime ships, and the network structure is relatively large and cannot be deployed to small devices. Therefore, there is an urgent need to provide a lightweight radar positioning method to ensure positioning accuracy and real-time performance. Summary of the Invention

[0004] The present invention provides a lightweight radar positioning and trajectory prediction method and device for the movement of maritime ships. The technical solutions are as follows:

[0005] On the one hand, an embodiment of the present invention provides a lightweight radar positioning method for the movement of maritime ships. The method includes:

[0006] Obtain a plurality of first samples; the first samples include known sample incident angles and their corresponding sample time-domain echo data;

[0007] Process the first samples to obtain second samples; the processing method is: preprocess the sample time-domain echo data according to the conventional beamforming method to obtain angle sample data, and generate labels for the angle sample data based on the known sample incident angles; wherein, the angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output;

[0008] Use the angle sample data as the input and the labels as the output to train the YOLO network with a plurality of second samples to obtain a trained YOLO network model;

[0009] Equivalent a maritime ship to a point target, and based on the time-domain echo data received by the radar and the pre-trained YOLO network, determine the distance and incident angle between the radar and the point target;

[0010] Use the determined distance and incident angle to determine the three-dimensional coordinates of the point target, and complete the radar positioning of the movement of the maritime ship.

[0011] On the other hand, an embodiment of the present invention provides a method for predicting the movement trajectory of a marine ship, including:

[0012] Using any of the above-mentioned lightweight radar positioning methods for marine ship movement, obtain the three-dimensional coordinates of the marine ship at a plurality of consecutive moments;

[0013] Using the three-dimensional coordinates at a plurality of consecutive moments and a pre-trained LSTM neural network, predict the predicted three-dimensional coordinates of the marine ship at the next moment.

[0014] On the other hand, a lightweight radar positioning device for marine ship movement is provided, and the device includes:

[0015] An acquisition unit for acquiring a plurality of first samples; the first samples include known sample incident angles and their corresponding sample time-domain echo data;

[0016] A processing unit for processing the first samples to obtain second samples; the processing method is: preprocess the sample time-domain echo data according to the conventional beamforming method to obtain angle sample data, and generate labels for the angle sample data based on the known sample incident angles; wherein, the angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output;

[0017] A training unit for using the angle sample data as input and the labels as output to train the YOLO network with a plurality of second samples to obtain a trained YOLO network model;

[0018] A determination unit for equivalent the marine ship to a point target, and based on the time-domain echo data received by the radar and the pre-trained YOLO network, determine the distance and incident angle between the radar and the point target;

[0019] A positioning unit for using the determined distance and incident angle to determine the three-dimensional coordinates of the point target, and complete the radar positioning of the marine ship movement.

[0020] On the other hand, a prediction device for the movement trajectory of a marine ship is provided, and the device includes:

[0021] An acquisition unit for using any of the above-mentioned lightweight radar positioning methods for marine ship movement to obtain the three-dimensional coordinates of the marine ship at a plurality of consecutive moments;

[0022] A prediction unit for using the three-dimensional coordinates at a plurality of consecutive moments and a pre-trained LSTM neural network to predict the predicted three-dimensional coordinates of the marine ship at the next moment.

[0023] On the other hand, a computer device is provided, which includes a memory and a processor, the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the above-mentioned lightweight radar positioning method for maritime ship movement and the method for predicting maritime ship movement trajectory.

[0024] On the other hand, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of the lightweight radar positioning method for the movement of a maritime ship and the method for predicting the movement trajectory of a maritime ship are implemented.

[0025] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned lightweight radar positioning method for maritime ship motion and the method for predicting maritime ship motion trajectory.

[0026] The technical solution provided by the present invention can at least bring the following beneficial effects:

[0027] The time-domain echo data received by the radar is preprocessed to obtain angle sample data, and labels of the angle sample data are generated based on the known sample incident angle. The angle sample data and labels obtained after preprocessing meet the input and output requirements of the YOLO network. The YOLO network is trained by taking the incident angle corresponding to the time-domain echo data as the classification category. The YOLO network does not need to use anchor frame detection, but only takes the angle and its category as output, which not only reduces the network structure, but also eliminates the NMS step. Its lightweight structure can not only be used in small devices, but also ensure positioning accuracy and real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 This is a flow chart of a lightweight radar positioning method for marine ship motion provided by one embodiment of the present invention;

[0030] Figure 2 This is a flow chart of a method for predicting the motion trajectory of a marine ship provided by an embodiment of the present invention;

[0031] Figure 3It is a structural diagram of a lightweight radar positioning device for the movement of a maritime ship provided by an embodiment of the present invention;

[0032] Figure 4 It is a structural diagram of a prediction device for the movement trajectory of a maritime ship provided by an embodiment of the present invention;

[0033] Figure 5 It is a hardware architecture diagram of a computer device provided by an embodiment of the present invention. Detailed implementation manners

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] Please refer to Figure 1 , a lightweight radar positioning method for the movement of a maritime ship provided by an embodiment of the present invention, the method includes:

[0036] Step 100, obtaining a plurality of first samples; the first samples include known sample incident angles and their corresponding sample time-domain echo data;

[0037] Step 102, processing the first samples to obtain second samples; the processing method is: preprocessing the sample time-domain echo data according to the conventional beamforming method to obtain angle sample data, and generating labels for the angle sample data based on the known sample incident angles; among them, the angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output;

[0038] Step 104, using the angle sample data as the input and the label as the output to train the YOLO network with a plurality of second samples to obtain a trained YOLO network model;

[0039] Step 106, equivalent the maritime ship to a point target, and based on the time-domain echo data received by the radar and the pre-trained YOLO network, determine the distance and incident angle between the radar and the point target;

[0040] Step 108, using the determined distance and incident angle to determine the three-dimensional coordinates of the point target, and completing the radar positioning of the movement of the maritime ship.

[0041] In an embodiment of the present invention, the time-domain echo data received by the radar is preprocessed to obtain angle sample data, and labels for generating the angle sample data are generated based on known sample incident angles. The angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output. By using the incident angle corresponding to the time-domain echo data as the classification category to train the YOLO network, the YOLO network does not need to use anchor boxes for detection and only outputs the angle and its category. This not only reduces the network structure but also eliminates the NMS step. Its lightweight structure can be applied not only to small devices but also ensure positioning accuracy and real-time performance.

[0042] The following describes Figure 1 the execution manners of the respective steps shown.

[0043] First, steps 100 "obtain a plurality of first samples", 102 "process the first samples to obtain second samples", and 104 "use the angle sample data as the input and the labels as the output to train the YOLO network with a plurality of second samples to obtain a trained YOLO network model" are described simultaneously.

[0044] During the navigation of a ship at sea, due to the influence of sea waves, its attitude angle changes complexly. In an embodiment of the present invention, based on the incident angle when the radar receives echo data and the distance between the radar and the ship, the three-dimensional coordinates of the ship's position can be calculated. Therefore, it is possible to consider using the time-domain echo data and the incident angle of the radar receiving echo data to train the YOLO network, so that it is not necessary to identify and detect the attitude of the ship through anchor boxes. Additionally, since the anchor box detection is omitted, the ship target can be equivalent to a point target, and the movement of the ship at sea is the continuous smooth curve movement of the point target.

[0045] In an embodiment of the present invention, the known sample incident angles and their corresponding sample time-domain echo data can be used as the first samples to implement the training of the YOLO network.

[0046] Among them, the FMCW (Frequency Modulated Continuous Wave) radar simulation parameters can be preset to achieve the positioning of ships at sea. For example, the simulation parameters are set as follows: the frequency is 24 GHz, the distance resolution is 1 m, the maximum detection distance is 220 m, the maximum detection speed is 30 m, an array antenna with 1 transmit and 8 receive, the antenna element spacing is half a wavelength, each Chkrp duration is 10 us, the Chkrp interval duration is 15 us, and the number of sampling points for each Chkrp is 256. At a known incident angle, the radar receives echo data every 10 ms, and each frame of received echo data is saved as a time-domain data matrix x(t) ∈ A^(K×L×A), where K = 256 represents the number of sampling points, L = 1 represents the number of Chkrps, and A = 8 represents the number of channels.

[0047] It should be noted that the incident angle is the azimuth angle at which the radar receives echo data. After determining the incident angle, echo data can be received based on this incident angle, and then corresponding time-domain echo data can be obtained.

[0048] When training the YOLO network, considering that the YOLO network has requirements for the matrix scale of the input and output, therefore, it is necessary to process the first sample so that the second sample obtained after processing meets the requirements of the YOLO network for the matrix scale of the input and output.

[0049] In one implementation, the requirement of the YOLO network for the scale of the input matrix is M*1. Therefore, when preprocessing the sample time-domain echo data according to the conventional beamforming method, it can be achieved in the following way:

[0050] When the range of the incident angle of interest is [-90°, 90°), the sample time-domain echo data is preprocessed according to the following formula:

[0051] P(θ) = Y·vec(Rxx) ∈ C 180×1

[0052]

[0053] Among them, P(θ) is the 180*1 angle sample data; vec() represents vectorizing the matrix column-wise, rvec() represents vectorizing the matrix row-wise, Rxx is the covariance matrix of the time-domain echo data x(t), a(θ) represents the antenna array direction vector, M represents the number of antenna elements, C is the complex space, and Y is an intermediate parameter;

[0054] Perform a zero-padding operation at the end of the angle sample data to obtain M*1 angle sample data; where M is an integer greater than 180.

[0055] Taking M = 192 as an example, since P(θ) is the 180*1 angle sample data, 12 zeros need to be supplemented at the end of the angle sample data to complete the zero-padding operation and obtain a 192*1 input matrix.

[0056] In one implementation, the YOLO network requires the scale of the output matrix to be k*n*q. Then, when generating the label of the angle sample data based on the known sample incident angle, it specifically may include:

[0057] Based on the M*1 matrix required as input by the YOLO network and the k*n*q matrix as output, the M*1 angular sample data obtained after zero-padding operation is divided into k equal sub-regions, and each sub-region includes the same number of micro-regions; the precision of each micro-region is the same as the precision required by the incident angle; the size of each sub-region is n; k, n, and q are all positive integers, and k*n = M;

[0058] For each angle in the angular sample data, the following operations are performed: determine the sub-region and micro-region to which the angle θ belongs, and use the sub-region and micro-region to which the angle θ belongs as the position to which the angle θ belongs. Generate the following label tar for the position to which the angle θ belongs θ , such that the label dimension is q-dimensional:

[0059]

[0060] Among them, if the angle is the incident angle corresponding to the x-th point target, then the classification label gx = 1; otherwise, the classification label gx = 0, where x = 1, 2, 3,..., m; q = m + 1, and m is a positive integer;

[0061] Generate a k*n*q label matrix based on the labels generated for each angle.

[0062] Since the number of ships at sea may be one or more, in order to be able to locate more ship point targets, when obtaining the first sample, there can be multiple ships at sea, so as to receive echo data at multiple incident angles corresponding to the multiple ships and form the first sample. In this way, the YOLO network trained using this sample can identify at most m ship point targets.

[0063] Furthermore, the sub-region i and micro-region j to which the angle θ belongs can be determined by the following formula:

[0064] i = Floor((θ + 90°) / n)

[0065] j = mod(θ + 90°, n)

[0066] When determining which sub-region the angle belongs to, the sub-region i can be calculated by rounding up using the above formula, and the modulo operation using the above formula is used to determine that the angle is in the j-th micro-region of the i-th sub-region.

[0067] Taking two point targets as an example, assume that the incident angle of point target 1 is θ 1 , and the incident angle of point target 1 is θ 2 , with 1° as a micro-region. Then, for the incident angle θ 1 of point target 1, the sub-region and micro-region to which it belongs are:

[0068] i = Floor(θ 1 + 90°) / n

[0069] j = mod(θ 1 + 90°, n)

[0070] For the incident angle θ of point target 1 1 The generated label is:

[0071]

[0072] For the incident angle θ of point target 2 2 Its sub-region and micro-region are:

[0073] i = Floor(θ 2 + 90°) / n

[0074] j = mod(θ 2 + 90°, n)

[0075] For the incident angle θ of point target 2 2 The generated label is:

[0076]

[0077] For example, assuming k = 8, n = 24, and taking the case where at most two point targets can be recognized, i.e., m = 2, q = 3, then for the angle range [-90°, 90°) of interest, the following 8 sub-regions can be divided:

[0078] [-90°, -66°), [-66°, -42°), [-42°, -18°), [-18°, 6°), [6°, 30°), [30°, 54°), [54°, 78°), [78°, 102°).

[0079] Since the scale of the last sub-region is different from that of other sub-regions, an expansion operation is required, that is, changing the scale of the last sub-region from [78°, 90°) to [78°, 102°).

[0080] Assuming the incident angle of point target 1 is 2° and the incident angle of point target 2 is 28°, then the position of the incident angle of point target 1 is: in the 20th micro-region of the 4th sub-region; the generated label is The position of the incident angle of point target 2 is: in the 22nd micro-region of the 5th sub-region; the generated label is

[0081] Based on this, in the 24*8*3 matrix, the element values at positions (4, 20, 1), (4, 20, 2), and (4, 20, 3) are 92, 1, and 0 respectively; the element values at positions (5, 22, 1), (5, 22, 2), and (5, 22, 3) are 112, 0, and 1 respectively. For other positions (x, y, z), when z = 1, the element value at the corresponding position is the sum of the corresponding angle and 90, and when z = 2 and 3, the element values at the corresponding positions are both 0.

[0082] From this, a 24*8*3 label matrix can be obtained.

[0083] That is to say, after generating the labels of the angle sample data based on the known sample incident angles, a k*n*q label matrix is obtained.

[0084] Therefore, the second sample includes an M*1 angle sample data matrix as the input and a k*n*q label matrix as the output; using multiple second samples to train the YOLO network, a trained YOLO network model can finally be obtained based on the set activation function.

[0085] Then, steps 106 "equivalent the maritime ship to a point target, and based on the time-domain echo data received by the radar and the pre-trained YOLO network, determine the distance and incident angle between the radar and the point target" and step 108 "use the determined distance and incident angle to determine the three-dimensional coordinates of the point target and complete the radar positioning of the maritime ship movement" are described simultaneously.

[0086] In practical applications, when using the trained YOLO network to perform radar positioning on the movement of a maritime ship, the maritime ship can be equivalent to a point target, and the time-domain echo data received by the radar is processed in the same way as the preprocessing of the first sample to obtain an M*1 angle sample data matrix corresponding to the time-domain echo data; by inputting the M*1 angle sample data matrix into the pre-trained YOLO network, a k*n*q label matrix can be obtained based on the output. According to the element values in the q dimension of the label matrix, determine which angles belong to which point targets, and then the incident angle can be obtained.

[0087] In an embodiment of the present invention, when determining the incident angle based on the time-domain echo data received by the radar and the pre-trained YOLO network, if there are two incident angles in the output label matrix that belong to different classification labels and these two incident angles belong to the same or adjacent micro-regions, then determine the different classification labels as the same point target, and determine the average value of the sum of these two incident angles as the incident angle of the same point target. In this case, the two different point targets identified are located in the same or adjacent micro-regions, indicating that the positions of the two point targets are very close and can be determined as one point target, thereby improving the accuracy of target recognition.

[0088] When determining the distance between the radar and the point target, the time-domain echo data matrix can be Fourier-transformed to obtain the distance between the radar and the point target.

[0089] After determining the distance and incident angle between the radar and the point target, the three-dimensional coordinates of the point target can be calculated in the following way:

[0090] Assume that the g coordinate system is the global coordinate system and the a coordinate system is the coordinate system of the radar receiving antenna device. The three-dimensional coordinates of the radar receiving antenna device are P t =(x t , y t , z t ), and the attitude coordinates are corresponding to the azimuth angle, pitch angle, and roll angle. The radar time-domain echo signal is received by the receiving antenna at the azimuth angle θ (incident angle), pitch angle distance d. The three-dimensional coordinates P t and the attitude coordinates of the receiving antenna device are A t , which are obtained through the sensors of the device itself.

[0091] First, use the following conversion formula to calculate the position of the time-domain echo data in the a - coordinate system:

[0092]

[0093] Then calculate the rotation matrix of the receiving antenna device

[0094]

[0095] Finally, obtain the position of the time-domain echo data in the g - coordinate system:

[0096]

[0097] Determine the position of the time-domain echo data in the global coordinate system as the three-dimensional coordinates of the ship point target, thus completing the radar positioning of the ship's movement at sea.

[0098] In the embodiment of the present invention, in the used YOLO network, there is no need to use anchor box detection, and only the angle and its category are used as the output, which not only reduces the network structure but also omits the NMS step. Its lightweight structure can not only be applied to small devices but also ensure the positioning accuracy and real-time performance.

[0099] Please refer to Figure 2 , the embodiment of the present invention provides a method for predicting the movement trajectory of a ship at sea, including:

[0100] Step 200: Using any of the lightweight radar positioning methods for the movement of maritime ships described above, obtain the three-dimensional coordinates of the maritime ship at several consecutive moments.

[0101] Step 202: Using the three-dimensional coordinates at several consecutive moments and the pre-trained LSTM neural network, predict the predicted three-dimensional coordinates of the maritime ship at the next moment.

[0102] Among them, the several consecutive moments are a set number of consecutive moments, and the consecutive moments at least include the current moment. When training the LSTM neural network, the three-dimensional coordinates at a set number of consecutive moments in the training samples are also used as the input, and the three-dimensional coordinates at the next moment are used as the output, so as to complete the training of the LSTM neural network.

[0103] Since the above embodiments can quickly and accurately locate the three-dimensional coordinates at the current moment, therefore, when predicting the three-dimensional coordinates at the next moment using the accurate three-dimensional coordinates at several consecutive moments, the prediction accuracy can also be improved, so as to realize the prediction of the movement trajectory of the maritime ship.

[0104] Please refer to Figure 3 , the embodiment of the present invention provides a lightweight radar positioning device for the movement of maritime ships. The device includes:

[0105] An acquisition unit 300, configured to acquire a plurality of first samples; the first samples include known sample incident angles and their corresponding sample time-domain echo data;

[0106] A processing unit 302, configured to process the first samples to obtain second samples; the processing method is: preprocess the sample time-domain echo data according to the conventional beamforming method to obtain angle sample data, and generate labels for the angle sample data based on the known sample incident angles; among them, the angle sample data and labels obtained after preprocessing meet the requirements of the YOLO network for input and output;

[0107] A training unit 304, configured to use the angle sample data as the input and the labels as the output to train the YOLO network using a plurality of second samples to obtain a trained YOLO network model;

[0108] A determination unit 306, configured to equivalent the maritime ship to a point target, and based on the time-domain echo data received by the radar and the pre-trained YOLO network, determine the distance and incident angle between the radar and the point target;

[0109] A positioning unit 308, configured to use the determined distance and incident angle to determine the three-dimensional coordinates of the point target, and complete the radar positioning of the movement of the maritime ship.

[0110] In one embodiment of the present invention, when the processing unit preprocesses the sample time-domain echo data according to the conventional beamforming method, it specifically includes:

[0111] When the range of the incident angle of interest is [-90°, 90°), the sample time-domain echo data is preprocessed according to the following formula:

[0112] P(θ) = Y·vec(Rxx) ∈ C 180×1

[0113]

[0114] where P(θ) is the angle sample data of 180*1; vec() represents vectorizing the matrix column-wise, rvec() represents vectorizing the matrix row-wise, Rxx is the covariance matrix of the time-domain echo data x(t), a(θ) represents the antenna array direction vector, M represents the number of antenna elements, C is the complex space, Y is an intermediate parameter;

[0115] Perform zero-padding at the tail of the angle sample data to obtain the angle sample data of M*1; where M is an integer greater than 180, and the YOLO network requires the input matrix to be of size M*1.

[0116] In one embodiment of the present invention, when the range of the incident angle of interest is [-90°, 90°) and the YOLO network can identify at most m ship point targets, when the processing unit generates the label of the angle sample data based on the known sample incident angle, it specifically includes:

[0117] Based on the M*1 matrix required for input by the YOLO network and the k*n*q matrix output, divide the M*1 angle sample data obtained after zero-padding into k equal sub-regions, and each sub-region includes the same number of micro-regions; the precision of the micro-regions is the same as the precision required by the incident angle; the size of each sub-region is n; k, n, and q are all positive integers, and k*n = M;

[0118] Based on each angle in the angle sample data, perform: determine the sub-region and micro-region to which the angle θ belongs, and use the sub-region and micro-region to which the angle θ belongs as the position to which the angle θ belongs, and generate the following label tar θ such that the label dimension is q-dimensional:

[0119]

[0120] where if the angle is the incident angle corresponding to the xth point target, the classification label gx = 1, otherwise, the classification label gx = 0, x = 1, 2, 3,..., m; q = m + 1; m is a positive integer;

[0121] Generate a k*n*q label matrix based on the labels generated for each angle.

[0122] In an embodiment of the present invention, when determining the incident angle based on the time-domain echo data received by the radar and the pre-trained YOLO network, if there are two incident angles in the output label matrix that belong to different classification labels and these two incident angles belong to the same or adjacent micro-regions, then determine the different classification labels as the same point target, and determine the average value of the sum of these two incident angles as the incident angle of the same point target.

[0123] Please refer to Figure 4 , an embodiment of the present invention provides a prediction device for the movement trajectory of a marine ship, and the device includes:

[0124] An acquisition unit 400, configured to use any one of the above-mentioned lightweight radar positioning methods for marine ship movement to acquire the three-dimensional coordinates of the marine ship at a plurality of consecutive moments;

[0125] A prediction unit 402, configured to use the three-dimensional coordinates at a plurality of consecutive moments and a pre-trained LSTM neural network to predict the predicted three-dimensional coordinates of the marine ship at the next moment.

[0126] It should be noted that: for the lightweight radar positioning device for marine ship movement and the prediction device for the movement trajectory of a marine ship provided in the above embodiments, only the above-mentioned division of each functional module is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the lightweight radar positioning device for marine ship movement provided in the above embodiments and the embodiment of the lightweight radar positioning method for marine ship movement belong to the same concept, and the prediction device for the movement trajectory of a marine ship provided in the above embodiments and the embodiment of the prediction method for the movement trajectory of a marine ship belong to the same concept. The specific implementation process can be seen in the method embodiments and will not be elaborated here.

[0127] An embodiment of the present application further provides a computer device. Please refer to Figure 5 , the computer device includes a processor and a memory. The memory stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, at least one program, the code set or the instruction set is loaded and executed by the processor to implement the lightweight radar positioning method for marine ship movement and the prediction method for the movement trajectory of a marine ship provided in the above method embodiments.

[0128] Embodiments of the present application also provide a computer-readable storage medium, on which at least one instruction, at least one program segment, a code set or an instruction set is stored. The at least one instruction, at least one program segment, the code set or the instruction set is loaded and executed by a processor to implement the lightweight radar positioning method for the movement of a maritime ship and the prediction method for the movement trajectory of the maritime ship provided in the above method embodiments.

[0129] Embodiments of the present application also provide a computer program product, which includes a computer program. The processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the lightweight radar positioning method for the movement of a maritime ship and the prediction method for the movement trajectory of the maritime ship described in any one of the above embodiments.

[0130] For convenience of description, when describing the above system or device, various modules or units are described separately according to functions. Of course, when implementing the present application, the functions of each unit can be implemented in one or more software and / or hardware.

[0131] From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0132] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0133] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A lightweight radar positioning method for marine ship motion, characterized in that: The method comprises: Acquire a plurality of first samples; the first samples include known sample incident angles and corresponding sample time domain echo data; The first sample is processed to obtain a second sample; the processing method is: preprocessing the sample time domain echo data according to a conventional beamforming method to obtain angle sample data, and generating a label of the angle sample data based on a known sample incident angle; wherein the angle sample data and the label obtained after the preprocessing meet the input and output requirements of the YOLO network; Taking the angle sample data as input and the label as output, training the YOLO network with the plurality of second samples to obtain a trained YOLO network model; Treat the ship at sea as a point target, and determine the distance and incident angle between the radar and the point target based on the time domain echo data received by the radar and the pre-trained YOLO network; The three-dimensional coordinates of the point target are determined using the determined distance and incident angle to complete the radar positioning of the movement of ships at sea.

2. The method according to claim 1, characterized in that The preprocessing of the sample time domain echo data according to the conventional beamforming method includes: When the incident angle range of interest is [-90°, 90°), the sample time domain echo data is preprocessed according to the following formula: P(θ)=Y·vec(Rxx)∈C 180×1 Wherein, P(θ) is the 180*1 angle sample data; vec() means vectorizing the matrix by columns, rvec() means vectorizing the matrix by rows, Rxx is the covariance matrix of the time domain echo data x(t), a(θ) represents the antenna array direction vector, M represents the number of antenna array elements, C is the complex space, and Y is the intermediate parameter; A zero padding operation is performed at the end of the angle sample data to obtain M*1 angle sample data; wherein M is an integer greater than 180, and the YOLO network requires the size of the input matrix to be M*1.

3. The method according to claim 2, characterized in that When the range of the incident angle of interest is [-90°, 90°), and the YOLO network can recognize at most m ship point targets, the label of the angle sample data generated based on the known sample incident angle includes: Based on the M*1 matrix required for input and the k*n*q matrix output by the YOLO network, the M*1 angle sample data obtained after the zero-padding operation is divided into k equal sub-regions, each of which includes the same number of micro-regions; the accuracy of the micro-regions is the same as the accuracy required by the incident angle; the size of each sub-region is n; k, n, q are all positive integers, and k*n=M; For each angle in the angle sample data, the following operations are performed: determining the sub-region and micro-region to which the angle θ belongs, taking the sub-region and micro-region to which the angle θ belongs as the position to which the angle θ belongs, and generating the following label tar for the position to which the angle θ belongs θ , so that the label dimension is q-dimensional: Among them, if the angle is the incident angle corresponding to the x-th point target, the classification label gx=1, otherwise, the classification label gx=0, x=1, 2, 3, ..., m; q=m+1; m is a positive integer; Based on the labels generated at each angle, a k*n*q label matrix is ​​generated.

4. The method according to claim 3, characterized in that When determining the incident angle based on the time domain echo data received by the radar and the pre-trained YOLO network, if there are two incident angles in the output label matrix that belong to different classification labels, and the two incident angles belong to the same or adjacent micro-areas, the different classification labels are determined as the same point target, and the mean of the sum of the two incident angles is determined as the incident angle of the same point target.

5. A method for predicting the motion trajectory of a marine vessel, characterized in that: include: Using the lightweight radar positioning method for the movement of a ship at sea as described in any one of claims 1 to 4 above, the three-dimensional coordinates of the ship at sea at several consecutive moments are obtained; The three-dimensional coordinates of several consecutive moments and the pre-trained LSTM neural network are used to predict the three-dimensional coordinates of the ship at sea at the next moment.

6. A lightweight radar positioning device for marine ship movement, characterized in that: The device comprises: An acquisition unit, configured to acquire a plurality of first samples; the first samples include known sample incident angles and corresponding sample time domain echo data; A processing unit is used to process the first sample to obtain a second sample; the processing method is: preprocessing the sample time domain echo data according to a conventional beamforming method to obtain angle sample data, and generating a label of the angle sample data based on a known sample incident angle; wherein the angle sample data and the label obtained after preprocessing meet the input and output requirements of the YOLO network; A training unit, configured to take the angle sample data as input and the label as output, so as to train the YOLO network using a plurality of second samples to obtain a trained YOLO network model; A determination unit is used to treat the maritime ship as a point target and determine the distance and incident angle between the radar and the point target based on the time domain echo data received by the radar and the pre-trained YOLO network; The positioning unit is used to determine the three-dimensional coordinates of the point target using the determined distance and incident angle, and complete the radar positioning of the movement of ships at sea.

7. A device for predicting the motion trajectory of a marine vessel, characterized in that: The device comprises: An acquisition unit, used to acquire the three-dimensional coordinates of the marine ship at a plurality of consecutive moments using the lightweight radar positioning method for marine ship motion as described in any one of claims 1 to 4 above; The prediction unit is used to predict the predicted three-dimensional coordinates of the ship at sea at the next moment by using the three-dimensional coordinates of several consecutive moments and the pre-trained LSTM neural network.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-5.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

10. A computer program product, characterized in that The method comprises a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 5.

Images