A radar signal multi-target tracking method and system based on a dynamic graph convolution network

Abstract

The application provides a radar signal multi-target tracking method and system based on a dynamic graph convolution network. In the application, a radar one-dimensional range image sequence of a target cluster is collected, and a radar echo two-dimensional sequence is formed through pulse compression; the position and amplitude data of overlapping scattering points are separated based on the scattering point distribution characteristics; the spatial distance between targets is calculated using the position information, a dynamic edge weight is constructed in combination with the radial velocity difference to generate a shielding relationship edge; the neighborhood features are aggregated through a dynamic graph convolution network to learn the track decoupling features of the shielding and intersection areas; the correlation weight is generated based on the decoupling features to input a multi-hypothesis tracking algorithm to generate multi-branch tracks, if the spatial distance of two branches is less than a dynamic threshold and the included angle of the velocity vectors exceeds a set angle, the branch with a higher correlation weight is retained, and a multi-target track sequence is output. The application significantly improves the multi-target tracking accuracy and robustness in the shielding and track intersection scenarios.

Description

Technical Field

[0001] This application relates to the field of radar signal processing technology, and in particular to a method and system for multi-target tracking of radar signals based on dynamic graph convolutional networks. Background Technology

[0002] In high-density target scenarios such as drone swarm tracking and vehicle platoon monitoring, radar systems must address the issue of scattering point aliasing caused by the small distance between targets, resulting in the loss of target location and identification information. Trajectory intersections cause track confusion, and traditional algorithms struggle to decouple overlapping features, leading to a sharp drop in tracking accuracy. Military and transportation applications require systems to maintain real-time response capabilities under strong clutter interference while ensuring trajectory continuity and identity consistency.

[0003] Currently, the mainstream approach employs a framework combining multi-hypothesis tracking (MST) and interactive multi-model filtering (IMF). The MST algorithm constructs trajectory hypothesis branches using distance and velocity correlation thresholds, selecting the trajectory scheme with the highest probability. The IMF runs multiple motion models in parallel, fusing the target's motion state based on model probability weighting, and dynamically updates the state estimate to adapt to the maneuvering target's behavior.

[0004] However, this scheme exhibits significant shortcomings in scenarios with high-density occlusion and trajectory intersections. Measurement loss due to occlusion leads to a surge in the number of associated hypotheses, resulting in heavy computational burden, frequent incorrect branch selections, and a significant exacerbation of trajectory confusion. Filters relying on pre-defined models struggle to respond to instantaneous target maneuvers, and the lag in motion state updates causes a substantial increase in position estimation bias. Summary of the Invention

[0005] This application provides a radar signal multi-target tracking method and system based on dynamic graph convolutional networks to solve the problem of high trajectory breakage rate in the prior art.

[0006] In a first aspect, this application provides a multi-target tracking method for radar signals based on dynamic graph convolutional networks, including:

[0007] A one-dimensional range image sequence of radar is acquired from the target cluster. The one-dimensional range image sequence is then pulse-compressed and arranged in a slow time dimension to form a two-dimensional radar echo sequence.

[0008] Based on the linear or curved distribution characteristics of the scattering points of the conical target, the position and amplitude of the aliased scattering points in the two-dimensional sequence of the radar echo are separated to obtain the scattering point position and amplitude data.

[0009] Based on the location information in the scattering point position and amplitude data, the spatial distance between targets is calculated, and dynamic edge weights are constructed by combining the radial velocity difference. When the spatial distance between targets is less than the radar resolution threshold, occlusion relationship edges are generated.

[0010] The node feature weights are initialized based on the amplitude information in the scattering point position and amplitude data. The dynamic edge weights are used as the neighborhood aggregation coefficients of the graph attention mechanism. The neighborhood node features of the occlusion relationship edges are aggregated through the dynamic graph convolutional network to learn the decoupling features of the target trajectory in the occlusion and intersection regions and generate trajectory decoupling features.

[0011] Based on the trajectory decoupling features, association weights are generated. The multi-hypothesis tracking algorithm is input to generate multi-branch trajectory hypotheses. Conflict resolution is performed based on the instantaneous velocity vector angle of the multi-branch trajectory hypotheses. When the spatial distance between two branches is less than the dynamic threshold and the velocity vector angle exceeds the set angle, the branch trajectory hypothesis with higher association weight is retained, and a multi-target trajectory sequence is output.

[0012] Optionally, node feature weights are initialized based on the scattering point location and amplitude information in the amplitude data. The dynamic edge weights are used as neighborhood aggregation coefficients for the graph attention mechanism. The neighborhood node features of the occlusion relationship edges are aggregated through a dynamic graph convolutional network to learn the decoupling features of the target trajectory in the occlusion and intersection regions, generating trajectory decoupling features, including:

[0013] Based on the amplitude information in the scattering point location and amplitude data, the amplitude value is converted into an initial node feature vector, and the node feature weights are initialized based on the initial node feature vector.

[0014] The dynamic edge weights are assigned to the neighborhood aggregation coefficients of the graph attention mechanism, and the initial node feature vector is subjected to a linear transformation operation through the learnable weight matrix to generate intermediate node features.

[0015] Using a dynamic graph convolutional network, with the occlusion relationship edges as connection paths, the features of neighboring nodes are aggregated by weighted summation of the intermediate node features of the neighborhood of the occlusion relationship edges.

[0016] The weighted summed features of the neighboring nodes are concatenated with the features of the intermediate nodes of the current node. A nonlinear activation operation is performed on the concatenated features to learn the decoupling features of the target trajectory in the occlusion and intersection regions. The updated feature vector is then output as the trajectory decoupling feature.

[0017] Optionally, based on the location information in the scattering point position and amplitude data, the spatial distance between targets is calculated, and a dynamic edge weight is constructed by combining the radial velocity difference. When the spatial distance between targets is less than the radar resolution threshold, an occlusion relationship edge is generated, including:

[0018] Based on the position information in the scattering point position and amplitude data, the spatial distance between any two scattering points is calculated, and the radial velocity difference between the scattering points is calculated according to the change of the position information in the slow time dimension.

[0019] The spatial distance between the targets is multiplied by the radial velocity difference, and dynamic edge weights are constructed based on the product result.

[0020] When the spatial distance between the targets is less than the radar resolution threshold, an occlusion relationship edge is generated between the corresponding scattering points, and the dynamic edge weight is used as the association strength value of the occlusion relationship edge.

[0021] Optionally, association weights are generated based on the trajectory decoupling features, and multi-branch trajectory hypotheses are generated by inputting a multi-hypothesis tracking algorithm. Conflict resolution is performed based on the instantaneous velocity vector angle of the multi-branch trajectory hypotheses. When the spatial distance between two branches is less than a dynamic threshold and the velocity vector angle exceeds a set angle, the branch trajectory hypothesis with the higher association weight is retained, and a multi-target trajectory sequence is output, including:

[0022] Based on the trajectory decoupling features, the similarity of feature vectors of different trajectory points is calculated to generate association weights, and the association weights are input into the multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses.

[0023] For the multi-branch trajectory hypothesis, when the spatial distance between any two branches is less than the dynamic threshold, the angle between the instantaneous velocity vectors of the two branches is further calculated. When the angle between the instantaneous velocity vectors exceeds the set angle, the correlation weights of the two branches are compared, the branch trajectory hypothesis with the higher correlation weight is retained, the retained branch trajectory hypotheses are integrated, and the multi-target trajectory sequence is output in chronological order.

[0024] Optionally, regarding the multi-branch trajectory hypothesis, when the spatial distance between any two branches is less than a dynamic threshold, the angle between the instantaneous velocity vectors of the two branches is further calculated. When the angle between the instantaneous velocity vectors exceeds a set angle, the correlation weights of the two branches are compared, and the branch trajectory hypothesis with the higher correlation weight is retained. The retained branch trajectory hypotheses are integrated, and a multi-target trajectory sequence is output in chronological order, including:

[0025] The multi-branch trajectory hypotheses whose spatial distance at the current moment is less than a dynamic threshold are selected to form a set of branch pairs to be processed, and the instantaneous velocity vector of each branch trajectory pair in the set of branch pairs to be processed is extracted at the current moment.

[0026] Calculate the angle between two instantaneous velocity vectors in the same branch trajectory pair, and obtain the angle value by vector dot product operation and inverse cosine function conversion;

[0027] When the included angle value exceeds the set angle, the association weight value of the two branch trajectories is extracted from the set of branch pairs to be processed; the association weight values ​​of the two branch trajectories are compared, the branch trajectory with the larger association weight value is retained, and the branch trajectory with the smaller association weight value is removed.

[0028] Integrate the retained branch trajectory assumptions and output the multi-target trajectory sequence in chronological order.

[0029] Optionally, based on the linear or surface distribution characteristics of the scattering points of the conical target, the positions and amplitudes of the aliased scattering points in the two-dimensional sequence of the radar echo are separated to obtain scattering point position and amplitude data, including:

[0030] Extract the position information of each scattering point in the two-dimensional radar echo sequence and its change sequence in the slow time dimension. Analyze the continuous migration pattern of the change sequence and identify whether the position distribution characteristics of the change sequence are linear or curved.

[0031] Based on the linear distribution characteristics of the scattering points of the cone target, and according to the continuity of the position offset direction and the smoothness of the amplitude change of adjacent slow time frames, scattering points that meet the continuity condition are assigned to the same target.

[0032] Based on the surface distribution characteristics of the scattering points of the cone target, the curvature change value of the scattering point position change sequence is calculated. When the curvature change value is less than a preset threshold and the position offset amplitude continues to increase, it is determined that the scattering points belong to the same target.

[0033] Based on the scattering points assigned to the same target, the spatial location information of overlapping scattering points is detected, and the positions and amplitudes of the aliased scattering points are separated to obtain the separated scattering point position and amplitude data.

[0034] Optionally, using a dynamic graph convolutional network, with the occlusion relationship edges as connection paths, the features of neighboring nodes are aggregated by weighted summation of the intermediate node features of the neighborhood of the occlusion relationship edges, including:

[0035] Based on the occlusion relationship edges, all neighboring nodes directly connected to the current node through the occlusion relationship edges are selected to form a set of neighboring nodes;

[0036] Extract the intermediate node features of each node in the neighborhood node set, wherein the intermediate node features are generated by linear transformation of the initial node features;

[0037] The dynamic edge weights are used as neighborhood aggregation coefficients. The intermediate node features of each neighborhood node are multiplied by the corresponding neighborhood aggregation coefficients to generate weighted neighborhood node features.

[0038] The features of all weighted neighboring nodes of the same current node are summed to output the aggregated neighboring node features.

[0039] Secondly, this application provides a radar signal multi-target tracking system based on a dynamic graph convolutional network, comprising:

[0040] The acquisition module is used to acquire the radar one-dimensional range image sequence of the target cluster, perform pulse compression processing on the one-dimensional range image sequence, and arrange it according to the slow time dimension to form a two-dimensional radar echo sequence.

[0041] The separation module is used to separate the position and amplitude of the aliased scattering points in the two-dimensional sequence of radar echoes based on the linear or curved distribution characteristics of the scattering points of the conical target, and to obtain the scattering point position and amplitude data.

[0042] The construction module is used to calculate the spatial distance between targets based on the position information in the scattering point position and amplitude data, and to construct dynamic edge weights by combining the radial velocity difference. When the spatial distance between targets is less than the radar resolution threshold, an occlusion relationship edge is generated.

[0043] The generation module is used to initialize node feature weights based on the amplitude information in the scattering point position and amplitude data, use the dynamic edge weights as the neighborhood aggregation coefficients of the graph attention mechanism, aggregate the neighborhood node features of the occlusion relationship edges through a dynamic graph convolutional network, learn the decoupling features of the target trajectory in the occlusion and intersection regions, and generate trajectory decoupling features.

[0044] The output module is used to generate association weights based on the trajectory decoupling features, input a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses, resolve conflicts based on the instantaneous velocity vector angle of the multi-branch trajectory hypotheses, and retain the branch trajectory hypothesis with higher association weight when the spatial distance between two branches is less than a dynamic threshold and the velocity vector angle exceeds a set angle, and output a multi-target trajectory sequence.

[0045] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a radar signal multi-target tracking method based on a dynamic graph convolutional network as described in the first aspect above.

[0046] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a radar signal multi-target tracking method based on a dynamic graph convolutional network as described in the first aspect.

[0047] This application acquires radar one-dimensional range image sequences of target clusters and performs pulse compression processing to form radar echo two-dimensional sequences. This step significantly improves the signal-to-noise ratio and resolution, laying the foundation for subsequent operations. Next, based on the scattering point distribution characteristics of the conical target, the position and amplitude data of the aliased scattering points are separated, effectively avoiding target position ambiguity and accurately extracting parameters. Then, based on the position information, the spatial distance between targets is calculated, and dynamic edge weights are constructed by combining radial velocity differences. When the distance is less than the radar resolution, occlusion relationship edges are generated to accurately quantify the relationship between targets and reduce trajectory confusion. Then, node feature weights are initialized based on amplitude information, and the dynamic edge weights are used as aggregation coefficients of the graph attention mechanism. The features of neighboring nodes are aggregated through a dynamic graph convolutional network to learn decoupling features, efficiently decoupling trajectory features in occlusion and intersection areas and solving the breakage problem. Finally, based on the decoupling features, associated weights are generated and input into a multi-hypothesis tracking algorithm. After generating the trajectory, conflict resolution is performed based on the velocity vector angle. When the spatial distance is small and the angle is large, the high-weight branch is retained to output the trajectory sequence, significantly reducing the false detection rate and identity switching rate.

[0048] Furthermore, based on the amplitude information of the scattering points, the amplitude values ​​are converted into initial node feature vectors and node feature weights are initialized, preserving the target's physical properties to prevent feature loss. Then, dynamic edge weights are assigned to the neighborhood aggregation coefficients of the graph attention mechanism, and the features are linearly transformed through a learnable weight matrix to generate intermediate node features, improving the adaptability to occlusion relationships. Subsequently, a dynamic graph convolutional network is used with occlusion relationship edges as connection paths to aggregate neighborhood node features through weighted summation, effectively compensating for feature transmission interruptions and maintaining trajectory continuity. Finally, the aggregated features and intermediate node features are concatenated and a nonlinear activation operation is performed to learn decoupling features, outputting an updated feature vector as the trajectory decoupling feature, significantly enhancing the trajectory recognition ability of occlusion and intersection regions and improving tracking accuracy.

[0049] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart of a radar signal multi-target tracking method based on a dynamic graph convolutional network provided in this application is shown;

[0052] Figure 2The illustration shows a scenario of a radar signal multi-target tracking method based on a dynamic graph convolutional network provided in this application.

[0053] Figure 3 A schematic diagram of the structure of a radar signal multi-target tracking system based on a dynamic graph convolutional network provided in this application is shown.

[0054] Figure 4 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0056] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0057] Researchers have found that existing methods in radar multi-target tracking have significant shortcomings in high-density occlusion and trajectory intersection scenarios: measurement loss due to occlusion leads to a surge in the number of associated hypotheses, resulting in heavy computational burden, frequent incorrect branch selections, and exacerbated trajectory confusion; simultaneously, motion filters relying on pre-defined models struggle to respond to instantaneous target maneuvers, and the lag in motion state updates leads to a significant increase in position estimation bias. For example, in dense target cluster environments, traditional methods cannot effectively separate aliased scattering points, exacerbating the risk of trajectory mismatch and position deviation. Therefore, a radar signal multi-target tracking method based on dynamic graph convolutional networks is urgently needed.

[0058] To address the aforementioned problems, this invention proposes a multi-target radar signal tracking method based on dynamic graph convolutional networks. Its core lies in dynamically modeling the occlusion relationships and intersection characteristics between targets using dynamic edge weights and an attention mechanism to achieve decoupled learning and efficient updating of trajectory features. Specifically, a one-dimensional radar range image sequence of a target cluster is acquired and pulse-compressed to form a two-dimensional sequence. The position and amplitude of aliased scattering points are separated based on the scattering point distribution characteristics of the conical target. Dynamic edge weights are constructed by calculating the spatial distance and radial velocity difference between targets based on position information. When the distance is less than the radar resolution threshold, occlusion relationship edges are generated. Node feature weights are initialized, and the dynamic edge weights are used as neighborhood aggregation coefficients for the graph attention mechanism. The decoupling characteristics of target trajectories in occlusion and intersection regions are learned by aggregating the neighborhood node features of occlusion edges through a dynamic graph convolutional network. Based on the decoupling characteristics, association weights are generated and input into a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses. The conflict between instantaneous velocity vector angles is resolved, and the branch trajectory with higher weight is retained to output the trajectory sequence. This method effectively solves the problem of a surge in associated assumptions in high-density occlusion and trajectory intersection scenarios, reduces computational burden and the frequency of erroneous branch selection, alleviates trajectory confusion, and improves motion state response speed through dynamic modeling and instant conflict resolution, avoids position estimation bias, and significantly enhances the robustness and accuracy of multi-target tracking.

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] Figure 1 This application provides a flowchart of a radar signal multi-target tracking method based on dynamic graph convolutional networks, as shown in the following embodiment. Figure 1 As shown, the method includes:

[0061] 101. Collect the radar one-dimensional range image sequence of the target cluster, perform pulse compression processing on the one-dimensional range image sequence, and arrange it according to the slow time dimension to form a two-dimensional radar echo sequence.

[0062] In the above steps, the target cluster refers to a collection of multiple target objects detected by the radar, such as a group of aircraft or vehicles; the radar one-dimensional range image sequence is a series of one-dimensional range data collected by the radar system in the slow time dimension. Each data point corresponds to a sequence of target range information received after a pulse transmission. This sequence is arranged in slow time order and includes the target's range distribution and amplitude; pulse compression processing is the process of filtering the radar echo signal using matched filter technology. By compressing the pulse width, the range resolution is improved, thereby distinguishing near-range targets; the slow time dimension represents the time scale of the radar pulse repetition interval (PRI), corresponding to the change in the time interval between different pulse transmissions; the radar echo two-dimensional sequence is a two-dimensional data structure formed after pulse compression processing. It is arranged in matrix form with slow time as the row index and range as the column index for subsequent signal analysis.

[0063] In this embodiment, the radar one-dimensional range image sequence of the target cluster is first acquired and pulse compression processing is performed to form a two-dimensional radar echo sequence. For example, in a radar system, a sequence of radar pulse signals, such as linear frequency modulated signals, is transmitted to the target cluster area. After each pulse is transmitted, the receiving system captures the echo signal and converts it into a digitized data sequence. Each sequence is called a one-dimensional range image. Assuming the system transmits 20 pulses, the echo sequence corresponding to each pulse contains 256 sampling points representing the values ​​of different range units. After acquisition, a sequence containing 20 range images of 256 points is formed. Then, pulse compression processing is applied to each one-dimensional range image, and a matched filter algorithm is used to improve the range resolution. Specifically, the received signal is converted to the frequency domain: first, a Fast Fourier Transform (FFT) is performed on the input signal to obtain the spectrum, and then it is multiplied with the spectrum of a pre-generated reference signal, such as the complex conjugate spectrum of the transmitted pulse. Here, the reference signal is obtained through sampling and numerical calculation. The range profile is generated by the matched filter output formula, which is Y(k) = X(k)·H*(k), where X(k) is the FFT spectrum of the received signal, H*(k) is the complex conjugate spectrum of the reference signal, and k is the frequency domain index. Finally, Y(k) is converted back to the time domain signal by the inverse fast Fourier transform (IFFT) to obtain the compressed range profile. For example, the echo sequence data of the first pulse is taken as [1.2, 0.8, 0.5], assuming the unit is amplitude value, and the reference signal spectrum H*(k) is assumed to be [0.9, 0.7, 0.6]. After FFT calculation, the spectrum is multiplied and then IFFT is used to return the compressed sequence as [1.1, 0.85, 0.55], which significantly improves the resolution. Finally, all the processed range profiles are rearranged according to the slow time dimension, i.e., the pulse transmission order. For example, 20 compressed 256-point sequences are combined into a 20-row, 256-column two-dimensional matrix, which is used as the radar echo two-dimensional sequence for signal processing in subsequent steps.

[0064] In a practical application, a radar detection experiment involved a scenario where a cluster of targets in the air obstructed each other and their trajectories intersected. Radar signal processing methods were employed to address this challenge. In the experiment, the radar was positioned at observation point A, pointing towards the airspace above region B, where multiple targets, such as targets C and D, were present. These targets obstructed each other during flight and their trajectories frequently intersected, making it difficult for a single radar measurement to discern target details. First, the radar continuously transmitted and received echo signals, acquiring a one-dimensional range profile sequence of the target cluster. Specifically, since the radar pulse repetition frequency was 10 kHz and the acquisition duration was 10 seconds, the total number of pulses was calculated as: repetition frequency × acquisition time = 10 kHz × 10 = 100 pulses. This yielded a one-dimensional range profile sequence corresponding to each pulse, containing signal distribution data in the range dimension. Then, pulse compression is performed on each one-dimensional range image sequence to improve the target's resolution in the range dimension. During processing, each input sequence has 200 sampling points, and a preset matched filter is used for calculation. For example, for each point in the sequence, the filter output value = the integral value of the input signal point × the reference signal point. The compressed high-resolution range image is generated through point-by-point calculation. Finally, these processed sequences are arranged along the slow time dimension, i.e., the time axis, and combined into a two-dimensional radar echo sequence: the vertical dimension represents the time step, with 100 time points corresponding to each pulse acquisition time, and the horizontal dimension represents the range sampling points, with 200 points in total. The whole sequence forms a 100×200 two-dimensional matrix, which facilitates subsequent analysis of the target's dynamic behavior under occlusion and intersecting trajectories.

[0065] In the overall scheme of step 101 above, by acquiring the radar one-dimensional range image sequence of the target cluster, the pulse compression technology is used to significantly improve the signal resolution and signal-to-noise ratio, effectively suppress interference sidelobes and enhance the target feature extraction capability; then, a structured radar echo two-dimensional sequence is constructed by precisely arranging it in the slow time dimension. This sequence completely preserves the coupled motion characteristics of the target in the radial range and slow time domain, providing a data basis with phase consistency for high-precision cluster target imaging, while greatly improving the subsequent motion trajectory analysis capability.

[0066] 102. Based on the linear or curved distribution characteristics of the scattering points of the conical target, separate the position and amplitude of the aliased scattering points in the two-dimensional sequence of the radar echo to obtain the scattering point position and amplitude data;

[0067] Optionally, step 102 may specifically include the following steps:

[0068] 1021. Extract the position information of each scattering point in the two-dimensional radar echo sequence and change the sequence in the slow time dimension. Analyze the continuous offset pattern of the change sequence and identify whether the position distribution characteristics of the change sequence are linear or curved.

[0069] 1022. Based on the linear distribution characteristics of the scattering points of the cone target, according to the continuity of the position offset direction of adjacent slow time frames and the smoothness of the amplitude change, the scattering points that meet the continuity condition are divided into the same target.

[0070] 1023. Based on the surface distribution characteristics of the scattering points of the cone target, calculate the curvature change value of the scattering point position change sequence. When the curvature change value is less than a preset threshold and the position offset amplitude continues to increase, determine that the scattering points belong to the same target.

[0071] 1024. Based on the scattering points assigned to the same target, detect the spatially overlapping position information, separate the position and amplitude of the aliased scattering points, and obtain the separated scattering point position and amplitude data.

[0072] In the above steps, a conical target refers to a cone-shaped structure being detected by radar, such as a missile warhead or similar device; a scattering point refers to a discrete location point on the target surface reflecting radar waves, each point containing position and amplitude information; linear distribution characteristics describe the linear trend of the scattering point position across different pulse emission times in the slow time dimension; curved distribution characteristics describe the curved path of the scattering point position change; aliased scattering points refer to scattering points where radar echoes overlap at the same location due to multiple targets being too close together, making them difficult to distinguish; the slow time dimension is a time scale measured in units of radar pulse repetition intervals, used to arrange pulse sequences; position information records the coordinate data of the scattering point in different slow time frames, such as distance values; amplitude data represents the scattering point. The concepts include: the intensity of the reflected signal; slow time frames, which refer to the complete data set corresponding to a single pulse echo, including multiple scattering points; the continuity of the position offset direction, meaning that the movement direction of scattering points remains consistent between adjacent slow time frames, whether positive or negative; the smoothness of amplitude changes, meaning that the amplitude values ​​of scattering points in adjacent frames fluctuate little and change continuously; the curvature change value, a numerical measure of the curvature of the position change path, measured in radians and calculated mathematically; the preset threshold, a critical value set to judge curvature stability or amplitude changes, such as a fixed number; spatial position overlap, meaning that scattering points from different targets appear at the same distance unit position; and the target, i.e., the individual object being detected. These concepts form the basis for subsequent processing of scattering point data.

[0073] In this embodiment, firstly, step 1021 extracts the position change sequence of each scattering point in the two-dimensional radar echo sequence and analyzes its distribution characteristics. The position change sequence of each scattering point is a continuous position value extracted from the slow time dimension. For example, in a slow time sequence, the position of the scattering point gradually moves from 150 meters in the first frame to 152 meters in the second frame, and then to 154 meters in the third frame. Next, the continuous offset pattern of the change sequence is analyzed, and the characteristics are determined by fitting the position points using the least squares method. If the fitted line is like the function y = at + b, where a and b are the residual values ​​of the fitting parameters, and the sum of the squares of the differences between the actual position and the fitted value is less than a preset threshold, such as 5% of the total distance, then the scattering point is considered to have a position change sequence. If the residuals are stable, the distribution is identified as linear; otherwise, a quadratic function is used to fit the curve. If the residuals are stable, the distribution is surface-like. For example, a scattering point position sequence is 150, 152, and 154 meters in three slow time frames. Using a linear fit, the function y = 2t + 148 (where t is the time index) is obtained. The residuals are calculated as follows: the actual value at 150 meters differs from the fitted value at 148 by 2 squared, which is 4; the difference at 152 meters is 0 squared, which is 0; and the difference at 154 meters is 0 squared, which is 0. The total residual value is 4. The threshold is set to 5% of the 4-meter difference in the position range from 150 to 154 meters, i.e., 0.2 meters. The residual of 4 is greater than the threshold of 0.2, so a quadratic function is used to fit the curve equation, such as y = 0.05t. 2 +1.9t+149, the new residual of 0.1 meters is less than the threshold of 0.2 meters, and is determined to be a curved distribution characteristic.

[0074] Secondly, based on the linear distribution characteristics of the scattering points in step 1022, they are classified into the same target; based on the continuity detection of the position offset direction, the position difference between adjacent slow-time frames is calculated, such as the offset from frame 1 to frame 2 being 152-150 = 2 meters (positive direction), and from frame 2 to frame 3 being 154-152 = 2 meters (positive direction). If the offset direction of three or more consecutive frames is consistent and all positive or all negative, for example, the positive movement sequence remains stable; at the same time, the smoothness of the amplitude change is checked, and the absolute percentage of the amplitude difference between adjacent frames is calculated, such as the amplitude of frame 1 being 0.8, frame 2 being 0.82, and so on. 3 frames 0.81, the change rate of the first to second frames is |0.82-0.8| / 0.8=2.5%, and the change rate of the second to third frames is |0.81-0.82| / 0.82≈1.22%. If the change rate is less than the threshold such as 5% and the condition is met for three consecutive frames, then these scattering points are determined to belong to the same target. For example, the sequence position change of target A is [150,152,154] meters (continuous positive direction) and the amplitude is [0.8,0.82,0.81] (fluctuation less than 5%), while the sequence of target B is [140,138,136] meters (negative direction movement) and is considered a different target.

[0075] Next, using step 1023, the scattering points based on the surface distribution characteristics are grouped into the same target; the curvature change value of the scattering point position change sequence is calculated, taking the coordinates of three consecutive points from the position sequence, such as frame 1 P1 (150 meters, amplitude 0.8), frame 2 P2 (152 meters, amplitude 0.82), and frame 3 P3 (155 meters, amplitude 0.83), where the position coordinates are considered as two-dimensional point vectors; then the curvature is calculated by dividing the vector cross product modulus by the modulus length, using the formula: Where Δr = P2 - P1 (coordinate difference vector, such as (2 meters, 0.02 amplitude)), Δr' = P3 - P2 (coordinate difference vector, such as (3 meters, 0.01 amplitude), the cross product modulus is the result of the cross operation of vector components, such as |2×0.01-0.02×3|=|0.02-0.06|=0.04 units per meter amplitude, and the cube of the modulus is the cube of the length of each vector; the specific process is: first calculate the vector difference, then calculate the cross product modulus, and finally divide by the cube of the modulus to obtain the curvature value; when five consecutive If the curvature values ​​of all frames are less than a preset threshold (e.g., 0.2 radians) and the position offset amplitude continues to increase continuously (e.g., the position value increases monotonically), then the scattering points are determined to belong to the same target. For example, taking three frames of data P1(150,0.8), P2(152,0.82), and P3(155,0.83), the vectors Δr=(2,0.02) and Δr'=(3,0.01) are calculated, and the cross product modulus=|2·0.01-0.02·3|=|0.02-0.06|=0.04. Unit: meter; Module length in cubic meters (2.0002) 3 ≈8.0024, curvature If the radian is less than the threshold of 0.2 and the position increases continuously from 150 to 155, it is classified as the same target, and similar sequences are checked repeatedly to confirm that five consecutive frames meet the conditions.

[0076] Finally, in step 1024, aliased scattering points are separated based on scattering points assigned to the same target, and their position and amplitude data are obtained. Points with overlapping spatial locations are detected within the group; for example, scattering points from different targets may be located at the same distance cell (e.g., 150 meters) in the same slow-time frame. Aliased points are separated by analyzing amplitude changes. When the amplitude of an overlapping point abruptly exceeds a smoothness threshold (e.g., a change exceeding 10% between adjacent frames), interpolation is used to reconstruct the true position. Specifically, at the same distance cell, if the amplitude value of adjacent frames suddenly increases (e.g., 0.8 at 150 meters in frame 1, and 1.6 in frame 2), it indicates that two scattering points are aliased. A cubic spline interpolation algorithm is then used to reconstruct the true position from the surrounding area. The location data is reconstructed into a curve, and the location offset is assigned to two points, 150.2 meters and 149.8 meters, with the amplitude divided according to the original ratio. For example, in the target group, the initial amplitude of the aliasing point is 1.6 at 150 meters. Assuming that the position and amplitude trends of the surrounding points show a reasonable segmentation ratio of 0.5:0.5, two scattering points are obtained after separation, with positions of 149.8 meters (amplitude 0.8) and 150.2 meters (amplitude 0.8), respectively. Finally, the separated scattering point position and amplitude matrix data are output. For example, for target 1, the output position sequence is [149.8, 151.9, ...] meters and the amplitude sequence is [0.8, 0.82, ...].

[0077] In practical applications, in a certain aerial target detection scenario, there is the issue of mutual obstruction and trajectory intersection of a group of targets in flight. The radar system uses an existing two-dimensional echo sequence, such as a 100×200 matrix output from step 101, where the vertical dimension represents 100 time steps and the horizontal dimension represents 200 range sampling points, to handle the problem of aliased scattering points and separate the characteristic information of different targets. Here, the change sequence of the position information of each scattering point in the two-dimensional sequence on the time axis is first extracted: for each scattering point in the sequence, its position offset value at all time steps is selected to form a sequence; for example, the position offset sequence of a typical scattering point may gradually change from position 10 units to position 50 units over time, for a total of 100 time steps. Next, the distribution characteristics of this change sequence are analyzed: if the change of the offset value shows a constant linear trend, such as the position value increasing or decreasing uniformly, it is judged as a linear distribution characteristic; otherwise, if the offset value has a curved or arc-shaped change, it is judged as a curved distribution characteristic. For linear distribution characteristics, the continuity of the offset direction between adjacent time steps in the sequence is checked. For example, the offset direction from time step 10 to step 11 should consistently increase positively, and the smoothness of the amplitude change should be checked, such as the amplitude value smoothly changing from 1.0 unit to 1.2 units. If these conditions are continuously met, the scattering point is classified into the same target group. For curved distribution characteristics, the curvature change value of the position change sequence is calculated. For example, curvature change value = |curvature at time step 50 - curvature at time step 25|, with a preset threshold of 0.1 units. If the calculated curvature change value is less than 0.1 and the position offset amplitude increases continuously, such as the displacement value increasing by an average of 0.5 units per step from time step 1 to 100, with a total displacement of 50 units, then the scattering point is classified as belonging to the same target. Finally, based on the above division results, the overlapping position information at a specific time step, such as time step 75, is detected. For example, two scattering points overlap near position 35 units. By adjusting the position and amplitude models, the aliased scattering points are separated, thus obtaining clear and separated position and amplitude data, which is convenient for subsequent target tracking.

[0078] In the overall scheme of step 102 above, based on the unique scattering point distribution characteristics of the conical target, a refined analysis of the two-dimensional radar echo sequence is performed. First, the scattering point position change sequence in the slow time dimension is extracted to accurately identify its linear or curved distribution pattern. For the linear distribution characteristics, the continuity of position offset between adjacent frames and the smooth change law of amplitude are used to determine the scattering point attribution. For the curved distribution characteristics, the curvature change threshold and position offset continuity constraint are used to confirm the target correlation. Finally, by detecting spatial position aliasing points and combining amplitude and position evolution characteristics, the scattering points are accurately separated, thereby obtaining scattering point position and amplitude data with clear physical meaning, spatiotemporal consistency and dealiasing, providing a clear underlying feature support for subsequent high-precision three-dimensional reconstruction and motion inversion.

[0079] 103. Based on the position information in the scattering point position and amplitude data, calculate the spatial distance between targets, and construct dynamic edge weights by combining the radial velocity difference. When the spatial distance between targets is less than the radar resolution threshold, generate occlusion relationship edges.

[0080] Optionally, step 103 may specifically include the following steps:

[0081] 1031. Based on the position information in the scattering point position and amplitude data, calculate the target spatial distance between any two scattering points, and at the same time, calculate the radial velocity difference of the scattering points according to the change of the position information in the slow time dimension.

[0082] 1032. Multiply the spatial distance between the targets with the radial velocity difference, and construct dynamic edge weights based on the product result;

[0083] 1033. When the spatial distance between the targets is less than the radar resolution threshold, an occlusion relationship edge is generated between the corresponding scattering points, and the dynamic edge weight is used as the association strength value of the occlusion relationship edge.

[0084] In the above steps, the spatial distance between targets represents the actual spatial interval between any two scattering points, calculated based on the position information of the scattering points, for example, using a distance formula to calculate the unit in meters; the position information is the coordinate data stored by the scattering points, including the position values ​​at different slow time points; the slow time dimension is a time scale with the radar pulse repetition interval as the unit, such as the time index of the pulse sequence; the radial velocity difference is the numerical difference between the velocities of two scattering points in the radar observation direction, calculated from the velocity change rate of the position information in the slow time dimension; the dynamic edge weight is the weight value of the edge in the constructed target relationship graph, obtained through mathematical operations, used to quantify the relationship strength; the product operation is an algorithm for multiplying two values, specifically using multiplication calculation; the radar resolution threshold is the minimum distance value at which the radar system can distinguish targets, such as a pre-set critical value like 0.5 meters; the obstruction relationship edge is the line in the radar target relationship graph representing obstruction between targets; the association strength value is the numerical weight stored in the obstruction relationship edge, reflecting the importance of the obstruction relationship; these concepts are used for subsequent processing of the interaction relationship between targets.

[0085] In this embodiment, firstly, step 1031 calculates the spatial distance between any two scattering points based on the position information in the scattering point position and amplitude data. Simultaneously, based on the changes in position information in the slow time dimension, the radial velocity difference between the scattering points is calculated. Specifically, the position sequence is extracted from the separated scattering point data. For example, the position sequence of a scattering point for target A is 100 meters in frame 1, 102 meters in frame 2, and 104 meters in frame 3; the position sequence of a scattering point for target B is 110 meters in frame 1, 112 meters in frame 2, and 104 meters in frame 3. Frame 114 meters; When calculating spatial distance, data from the same slow-time frame is used, and the Euclidean distance formula is applied: d = |x1 - x2|, where x1 and x2 are the position values ​​(in meters) of the two scattering points at the same time point. For example, in frame 1, the position difference between the scattering points of target A and target B is |100 - 110| = 10 meters. When calculating the radial velocity difference, the rate of change of velocity position for each scattering point is first calculated. For example, the velocity of the scattering point of target A is calculated as the position sequence difference divided by the time interval. Assuming the pulse interval is a fixed value of 0.1 seconds, the formula is... Where Δx is the position change value, Δt is the slow-time frame interval (in seconds), the position difference between target A in frames 1 and 2 is 102 - 100 = 2 meters, the time interval is 0.1 seconds, and the speed is... m / s, the velocity of the scattering point of target B can be obtained using the same method. m / s; then calculate the radial velocity difference, formula Δv=|v A -v B For example, the difference |20-20| = 0 m / s; the final output is the distance and velocity difference matrix of all scattering point pairs. For example, the result of target AB pairing: spatial distance 10 m, radial velocity difference 0 m / s.

[0086] Secondly, in step 1032, the spatial distance between the targets and the radial velocity difference are multiplied, and a dynamic edge weight is constructed based on the product result. Specifically, from the data output in step 1031, the calculated spatial distance and radial velocity difference for each scattering point pair are taken, and a direct multiplication operation is performed using the formula w = d × |Δv|, where w is the dynamic edge weight value (unit: meters). 2 d is the spatial distance (in meters), |Δv| is the absolute value of the velocity difference (in meters per second), and the product value is used to quantify the strength of the dynamic relationship between targets; for example, from the previous example of target AB pairing: distance 10 meters, velocity difference 0 meters per second, calculate w = 10 × |0| = 0 meters. 2 / second indicates a weight value of 0; for example, if another paired target CD is 5 meters away and has a speed difference of 2 meters / second, the calculated w = 5 × 2 = 10 meters. 2 / second, output weight value 10; during construction, repeat this multiplication for all scattering point pairs and store the weight values ​​as a dynamic edge weight matrix for subsequent steps.

[0087] Finally, in step 1033, when the spatial distance between the targets is less than the radar resolution threshold, an occlusion relationship edge is generated between the corresponding scattering points, and the dynamic edge weight is used as the association strength value of the occlusion relationship edge. Specifically, the spatial distance between each pair of scattering points is compared with a preset radar resolution threshold, for example, 0.5 meters. If the distance is less than 0.5 meters, an occlusion relationship edge is generated, representing that the two targets will occlude each other. Simultaneously, the weight value of the pair is taken from the dynamic edge weight and directly assigned to this edge as the association strength value, reflecting the dynamic influence of the occlusion relationship. For example, if the spatial distance between targets A and B is 10 meters, which is greater than the threshold of 0.5 meters, no edge is generated; while the distance between paired targets E and F is 0.3 meters, which is less than the threshold of 0.5 meters, a connecting line from target E to target F is generated to represent the occlusion relationship, and the previously calculated dynamic edge weight value for the pair is taken, such as weight w = 8 meters from the output of step 1032. 2 / second, assigned as the association strength value; the final output relationship graph structure, in which the number and strength of edges are automatically constructed based on the above rules. For example, in a scenario with 10 scattering points, if 5 pairs satisfy the distance is less than the threshold, 5 edges are added along with the association strength data.

[0088] In practical applications, in a radar monitoring scenario, to address the issues of obstruction and trajectory intersection of a group of flying targets in the air, dynamic analysis of the relationships between targets is performed based on separated scattering point data, including the position and amplitude information of 100 target scattering points, covering 100 time steps. For example, at time step 50, the scattering point position of target A is (30, 40, 50) units, and the scattering point position of target B is (32, 38, 48) units. First, the spatial distance between these two scattering points is calculated using the formula: Substitute specific values ​​to calculate Units. Simultaneously, the radial velocity difference is calculated based on the change in position over time: Target A changes from position (20,30,40) units at time step 30 to position (30,40,50) units at time step 50, with a change time of 20 steps (0.1 seconds per step, totaling 2 seconds). The radial velocity is calculated as follows: The unit is 7.5 units / second; similarly, the radial velocity of target B is 7.5 units / second, therefore the velocity difference is |8.66 - 7.5| = 1.16 units / second. Next, dynamic edge weights are constructed: multiplying the spatial distance of 3.46 units by the velocity difference of 1.16 units / second, the calculation is: weight = 3.46 × 1.16 ≈ 4.01 units. 2 / second. Finally, the occlusion relationship is determined. A radar resolution threshold of 0.5 units is preset. During a time step of 40 to 60 seconds, if the spatial distance between the two scattering points is consistently less than the threshold (e.g., between 0.1 and 0.3 units), an occlusion relationship edge is generated between the scattering points, with a dynamic edge weight of 4.01 units. 2 / second, as the correlation strength value, is used to describe the dynamic impact of targets during mutual occlusion.

[0089] In the overall scheme of step 103 above, by accurately analyzing the location and amplitude data of scattering points, the spatial distance between targets is first calculated and the radial velocity difference is extracted simultaneously. The distance and velocity difference are innovatively multiplied to construct dynamic edge weights, and the weight values ​​can respond to changes in the target motion state in real time. When the target spacing exceeds the radar resolution threshold, an occlusion relationship edge with correlation intensity quantification is automatically generated. This intensity value integrates the target spatial position constraints and relative motion characteristics, effectively solving the problem of misjudgment of moving targets in traditional distance judgment. It forms a multi-dimensional fusion feature that includes the target interaction intensity and the dynamic level of occlusion risk, providing key data support with physical meaning and time adaptability for subsequent cluster target topology construction and occlusion compensation algorithms.

[0090] 104. Initialize node feature weights based on the amplitude information in the scattering point position and amplitude data, use the dynamic edge weights as the neighborhood aggregation coefficients of the graph attention mechanism, aggregate the neighborhood node features of the occlusion relationship edges through a dynamic graph convolutional network, learn the decoupling features of the target trajectory in the occlusion and intersection regions, and generate trajectory decoupling features.

[0091] Optionally, step 104 may specifically include the following steps:

[0092] 1041. Based on the amplitude information in the scattering point location and amplitude data, the amplitude value is converted into an initial node feature vector, and the node feature weights are initialized based on the initial node feature vector;

[0093] 1042. Assign the dynamic edge weights to the neighborhood aggregation coefficients of the graph attention mechanism, and perform a linear transformation operation on the initial node feature vector through the learnable weight matrix to generate intermediate node features;

[0094] 1043. Using a dynamic graph convolutional network, with the occlusion relationship edges as connection paths, the features of the neighboring nodes are aggregated by weighted summation of the intermediate node features of the neighborhood of the occlusion relationship edges.

[0095] Specifically, step 1043 may include the following process: based on the occlusion relationship edges, select all neighboring nodes directly connected to the current node through the occlusion relationship edges to form a neighboring node set; extract the intermediate node features of each node in the neighboring node set, wherein the intermediate node features are generated by linear transformation of the initial node features; use the dynamic edge weights as neighborhood aggregation coefficients, multiply the intermediate node features of each neighboring node by the corresponding neighborhood aggregation coefficients to generate weighted neighborhood node features; sum all the weighted neighborhood node features of the same current node to output aggregated neighborhood node features.

[0096] 1044. The weighted summed features of the neighboring nodes are concatenated with the features of the intermediate nodes of the current node. A nonlinear activation operation is performed on the concatenated features to learn the decoupling features of the target trajectory in the occlusion and intersection regions. The updated feature vector is then output as the trajectory decoupling feature.

[0097] In the above steps, amplitude information refers to the echo signal intensity values ​​recorded in the scattering point data obtained in step 102; the initial node feature vector is a multi-dimensional numerical array transformed from amplitude values ​​through linear mapping, used to represent the initial feature state of the target; node feature weights are numerical parameters assigned to each scattering point, representing its importance in the network; the graph attention mechanism is an artificial intelligence model that can automatically focus on important nodes; the neighborhood aggregation coefficient is the weight ratio value used when aggregating neighbor node information; the dynamic edge weight is the relationship strength value output from step 103; the dynamic graph convolutional network is a neural network model that can handle dynamic connection relationships; the learnable weight matrix is ​​a transformation matrix that is automatically adjusted through training; the linear transformation operation is an algorithm that multiplies the input vector with a matrix; and the intermediate node features are the node features generated after the linear transformation. Vector; Neighborhood node features, which are feature vectors of nearby nodes connected by occlusion relationship edges; Weighted neighborhood node features, which are the results of multiplying neighborhood node features by coefficients; Target trajectory, which refers to the target's movement path over time; Occlusion and intersection regions, which describe the spatial position of mutual influence between targets; Decoupling features, which are independent target features separated after model learning; Trajectory decoupling features, which are the final output feature vectors representing the target path; Connection path, which refers to the connection relationship between nodes formed by occlusion relationship edges; Neighborhood node set, which is the set of all nodes directly connected to the current node; Weighted summation, which is the process of multiplying multiple features by weights and then summing them; Concatenation, which is the operation of concatenating multiple feature sequences into a new sequence; Nonlinear activation operation, which is the step of applying a nonlinear function to the input values ​​and the output; Updated feature vector, which is the new feature sequence output after network processing.

[0098] In this embodiment, firstly, step 1041 converts the amplitude value into an initial node feature vector based on the amplitude information in the scattering point position and amplitude data, and initializes the node feature weights. The specific process is as follows: extract the amplitude value from the separated scattering point amplitude data. For example, if the amplitude of a scattering point of target A is 0.8, convert it into a fixed-dimensional vector, assuming the dimension is 2. The conversion formula is direct mapping. For example, the amplitude value a is converted into the vector [a,0], resulting in the initial node feature vector. For example, the feature vector of target A is [0.8,0]. Then, initialize the node feature weights and assign a preset initial value, such as 0.5 for all nodes, representing their basic importance in feature learning. For example, if the scattering point amplitudes of targets A and B are 0.8 and 1.2 respectively, the initial feature vectors after conversion are [0.8,0] and [1.2,0] respectively, and the weights of all nodes are initialized to 0.5.

[0099] Secondly, in step 1042, dynamic edge weights are assigned to the neighborhood aggregation coefficients of the graph attention mechanism, and a linear transformation operation is performed on the initial node feature vector using a learnable weight matrix to generate intermediate node features. Specifically, the process is as follows: first, values ​​are taken from the dynamic edge weight matrix output in step 103, such as a weight of 0.5 for the pairing of targets A and B, and stored as neighborhood aggregation coefficients; then, a learnable weight matrix, such as a randomly initialized matrix W of assumed size 2×2, is applied, and the linear transformation operation formula F is used. mid =W·F initial , where F initial Here, W is the initial node feature vector, and F is the weight matrix. mid These are intermediate node features; for example, the initial features of target A are [0.8, 0]. Assuming the weight matrix W = [[1.0, 0.5], [0.2, 1.0]], calculate F... mid =[1.0*0.8+0.5*0,0.2*0.8+1.0*0]=[0.8,0.16], the output is the feature vector of the intermediate node [0.8,0.16].

[0100] Next, in step 1043, using the occlusion relationship edges as connection paths, the features of neighboring nodes are aggregated through a dynamic graph convolutional network. Specifically, based on the occlusion relationship edges generated in step 103 (e.g., there is an edge between targets A and B), the neighboring nodes directly connected to the current node are selected to form a set of neighboring nodes (e.g., the current node is A, and the neighbor is B). The intermediate node features of each node in this set are extracted, such as the features of B [1.44, 0.24] taken from the output of step 1042. Then, the dynamic edge weights are used as neighborhood aggregation coefficients, such as the weight of the AB edge. 0.5 is used to multiply the features of each neighboring node by a coefficient to generate weighted neighboring node features. For example, the feature of B [1.44, 0.24] is multiplied by 0.5 to get [0.72, 0.12]. Finally, the weighted neighboring node features are summed. If there is only one neighbor, B, then [0.72, 0.12] is directly output as the aggregated neighboring node features. For example, in a graph containing nodes A, B, and C, the neighborhood set of the current node A is B, the weight of the AB edge is 0.5, and the aggregated features are directly taken as the weighted result [0.72, 0.12].

[0101] Finally, in step 1044, the weighted summed features of the neighboring nodes are concatenated with the features of the intermediate nodes of the current node. A nonlinear activation operation is performed on the concatenated features to learn the decoupling features of the target trajectory in the occlusion and intersection regions, and the trajectory decoupling features are output. The specific process is as follows: First, the features of the intermediate nodes of the current node, such as the features of A output from step 1042 [0.8, 0.16], and the aggregated features of the neighboring nodes from step 1043, such as [0.72, 0.12], are concatenated into a new vector [0.8, 0.16, 0.72, 0.12]. Then, a nonlinear activation operation, such as the ReLU function formula max(0, x), is applied to process each element independently. For example, if 4 elements are input, 4 elements are output. For example, if the first element is 0.8, it will still be 0.8 after ReLU. All elements are greater than 0 and there is no change. The output is used as the trajectory decoupling feature, representing the separation feature path information of the target in the occlusion region. Finally, a feature vector such as [0.8, 0.16, 0.72, 0.12] is generated.

[0102] In practical applications, in a radar tracking scenario, facing the signal obstruction problem caused by the intersection of the trajectories of multiple flying targets, dynamic graph convolution is used to decouple trajectory features based on the scattering point data generated in step 103, which includes the position, amplitude, and obstruction relationship edges of 100 scattering points. Specifically, the amplitude value is first converted into node features: for example, the amplitude value of scattering point A is 3.2 units, which is converted into an initial feature vector [1.5, -0.3] through a fully connected layer. This feature is stored in memory as a 32-bit floating-point number, occupying 8 bytes of space. Secondly, based on the dynamic edge weights calculated in step 103, such as the 4.01 unit weight between targets A and B... 2 / second is used as the neighborhood aggregation coefficient, and the initial features are linearly transformed: Let the learnable weight matrix be a 2×2 matrix [[0.8,-0.2],[0.1,0.6]], and calculate the intermediate features of point A = [1.5×0.8+(-0.3)×(-0.2),1.5×0.1+(-0.3)×0.6] = [1.26,-0.03]. Next, neighboring nodes are selected: When point A is connected to points B and C through occlusion edges (the intermediate features of point B are [0.9, 0.4], and those of point C are [-0.5, 1.2]), the features are weighted using edge weights: AB edge weight is 4.01, so the weighted feature of point B = [0.9 × 4.01, 0.4 × 4.01] = [3.61, 1.60]; AC edge weight is 2.3, so the weighted feature of point C = [-0.5 × 2.3, 1.2 × 2.3] = [-1.15, 2.76]. The aggregated neighborhood feature = [3.61 + (-1.15), 1.60 + 2.76] = [2.46, 4.36]. Finally, the aggregated feature is concatenated with the intermediate feature of point A itself [-0.03] to obtain [2.46,4.36,-0.03]. After ReLU activation, the threshold is 0, and negative values ​​are zeroed out, and the final trajectory decoupling feature [2.46,4.36,0] is output. This three-dimensional feature vector describes the motion characteristics of the target in the trajectory intersection area.

[0103] In the overall scheme of step 104 above, by innovatively fusing scattering point amplitude information and dynamic occlusion relationship features, the scattering point amplitude value is first converted into initial node feature weights to achieve a preliminary characterization of the target's physical properties. Then, the dynamic edge weights are precisely injected into the neighborhood aggregation coefficients of the graph attention mechanism. The initial features are linearly transformed by a learnable weight matrix to generate intermediate node features containing spatial correlation. Subsequently, the neighborhood node set is dynamically selected based on the occlusion relationship edges, and the intermediate features of the neighborhood nodes are weighted and summed using the dynamic edge weights as coefficients, creatively realizing the fusion and transmission of occlusion region features. Finally, by splicing the current node and neighborhood aggregation features and applying nonlinear activation, the spatiotemporal evolution characteristics of the target in the intersection region of occlusion and trajectory are deeply mined, and a breakthrough trajectory decoupling feature with strong discriminative power is generated. This feature not only retains the morphological and structural information of the target in the occlusion state, but also effectively decouples the trajectory confusion effect, providing a key discrimination basis for accurate multi-target identification in complex scenes.

[0104] 105. Generate association weights based on the trajectory decoupling features, input the multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses, resolve conflicts based on the instantaneous velocity vector angle of the multi-branch trajectory hypotheses, and retain the branch trajectory hypothesis with higher association weight when the spatial distance between two branches is less than the dynamic threshold and the velocity vector angle exceeds the set angle, and output the multi-target trajectory sequence.

[0105] Optionally, step 105 may specifically include the following steps:

[0106] 1051. Based on the trajectory decoupling features, calculate the feature vector similarity of different trajectory points, generate association weights, and input the association weights into the multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses;

[0107] 1052. Regarding the multi-branch trajectory hypothesis, when the spatial distance between any two branches is less than the dynamic threshold, the angle between the instantaneous velocity vectors of the two branches is further calculated. When the angle between the instantaneous velocity vectors exceeds the set angle, the correlation weights of the two branches are compared, the branch trajectory hypothesis with the higher correlation weight is retained, the retained branch trajectory hypotheses are integrated, and the multi-target trajectory sequence is output in chronological order.

[0108] Step 1052 may specifically include the following processes: filtering the multi-branch trajectory hypotheses whose spatial distance at the current moment is less than a dynamic threshold to form a set of branch pairs to be processed; extracting the instantaneous velocity vector of each branch trajectory pair in the set of branch pairs to be processed at the current moment; calculating the angle between the two instantaneous velocity vectors in the same branch trajectory pair, the angle being obtained by vector dot product operation and inverse cosine function conversion; when the angle exceeds a set angle, extracting the correlation weight values ​​of the two branch trajectories from the set of branch pairs to be processed; comparing the magnitude of the correlation weight values ​​of the two branch trajectories, retaining the branch trajectory with the larger correlation weight value, and removing the branch trajectory with the smaller correlation weight value; integrating the retained branch trajectory hypotheses and outputting the multi-target trajectory sequence in chronological order.

[0109] In the above steps, the trajectory decoupling feature is the feature vector data representing the independent spatial and temporal motion of the target, output from step 104, containing position and velocity information; the association weight is a value calculated by comparing the similarity of feature vectors of different trajectory points, representing the strength of the correlation between trajectory points; the multi-hypothesis tracking algorithm is a standard tracking method that generates multiple possible target motion path schemes; the multi-branch trajectory hypothesis is the multiple trajectory path branches generated by the algorithm, each branch representing a complete trajectory sequence of a potential target; the instantaneous velocity vector angle is the angle of difference between the velocity directions of two branches at the same time point, calculated by vector dot product and inverse cosine function; conflict resolution is the process of handling branch overlap, used to eliminate unreasonable trajectory paths; the dynamic threshold is a distance threshold that changes with time or scene, used to determine whether the distance between branches is small enough; and the set angle is a pre-defined... The standard angle value, such as 90 degrees, serves as the boundary for judging differences in velocity direction; the branch trajectory hypothesis with higher association weight indicates a more reliable trajectory branch retained in conflict resolution; the multi-target trajectory sequence is the trajectory data of multiple targets in the time series of the final output; a trajectory point is a single location point in the trajectory path; feature vector similarity is a quantified value of the degree of difference between the feature vectors of trajectory points calculated by mathematical methods such as cosine similarity; the branch trajectory pair set is a set of pairs composed of multiple trajectory branches; the unprocessed branch pair set is a set of conflicting branch pairs that meet the distance condition at the current time point; the instantaneous velocity vector is the directional quantity composed of the velocity direction and magnitude of the branch at a certain time point; the vector dot product operation is a mathematical formula for multiplying and summing two vectors; the inverse cosine function is a mathematical function for calculating angle values; the included angle value is the calculated specific angle value used to judge differences in velocity direction.

[0110] In this embodiment, firstly, step 1051 calculates the feature vector similarity of different trajectory points based on trajectory decoupling features to generate association weights, and then inputs these weights into a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses. Specifically, the process involves extracting the feature vector of each trajectory point from the trajectory decoupling features. For example, the feature vector of target 1 at time t1 is [0.8, 0.16, 0.72, 0.12]. The similarity between any two points is calculated using the cosine similarity formula. Where F i and F jThese are the feature vectors (unitless) of trajectory points i and j. The dot product operation is the sum of the element-wise multiplication of the vectors, and the magnitude is the Euclidean norm of the vector. For example, if point A has features [0.8, 0.16, 0.72, 0.12] and point B has features [0.9, 0.18, 0.75, 0.13], the dot product is 0.8*0.9 + 0.16*0.18 + 0.72*0.75 + 0.12*0.13 = 0.72 + 0.0288 + 0.54 + 0.0156 ≈ 1.3044, and the magnitude is... Similarly, ||F j The similarity is approximately 1.115, calculated as 1.3044 / (1.095*1.115)≈1.3044 / 1.221≈1.068. This is normalized to a weight of 0.8 (0.8 represents 0.8). The similarity value is then directly converted into association weights; for example, a similarity of 1.068 between points A and B is mapped to a weight of 0.8. All point-to-point weights are combined into a matrix. Next, a multi-hypothesis tracking algorithm, such as the JPDA algorithm, is input. This algorithm generates multiple trajectory branches based on the weights. For example, it assumes trajectory branch 1 represents a possible path for target A, branch 2 represents a possible path for target B, and outputs a multi-branch trajectory hypothesis sequence, such as the branch 1 position sequence [100 meters, 102 meters, ...], and the branch 2 sequence [110 meters, 112 meters, ...]. For example, in a radar tracking scenario with 10 trajectory points, this calculation generates a weight matrix, which is then input into the algorithm, outputting 3 branch trajectories.

[0111] Secondly, conflict resolution is performed based on the multi-branch trajectory hypothesis in step 1052. When the spatial distance between two branches is less than the dynamic threshold and the angle between their velocity vectors exceeds a set angle, the correlation weights are compared and the branch trajectory hypothesis with the higher weight is retained. The multi-target trajectory sequence is then integrated and output. The specific process is as follows: First, multi-branch trajectory hypotheses whose spatial distance at the current time point is less than the dynamic threshold are selected. For example, the dynamic threshold is calculated based on radar resolution and velocity, with an initial value of 0.5 meters, which is automatically adjusted over time to form a set of branch pairs to be processed. For example, the distance between branch 1 and branch 2 at time t is... The distance between |100m-102m| = 2m, but the dynamic threshold is 2.5m, which is not satisfied; the distance between branches 3 and 4 is |150m-150.3m| = 0.3m, which is less than the threshold of 0.5m, so it is added to the set); then, extract the instantaneous velocity vector of the branch pair in the set at the current moment (e.g., take the velocity vector of branch 3 at ± [20m / s, 5m / s] from the assumed branch trajectory data, representing the components of the 20m / s direction and the 5m / s amplitude); calculate the angle between the two velocity vectors of the same branch pair, using the dot product formula, v1·v2 = v 1x v 2x +v 1y v 2y , module length Angle formula Among them, the inverse cosine function cos -1 Output angle values ​​(in degrees), for example, the dot product of the velocity vectors of branch 3 [20, 5] and branch 4 [10, 0] is 20*10 + 5*0 = 200, and the magnitude is... included angle θ = cos -1 (200 / (20.62×10))≈cos -1 (200 / 206.2)≈cos -1 (0.97)≈14 degrees is less than the set angle of 90 degrees; if it exceeds the set angle, such as branch 5 velocity [0,30] and branch 6 velocity [30,0], dot product 0, modulus 30 each, included angle θ=cos-1(0)=90 degrees equal to the preset value of 90 degrees, then the associated weights are extracted from the weight matrix output in step 1051, such as branch 5 weight 0.7 and branch 6 weight 0.4; compare the weight values, retain the branch trajectory hypothesis with higher weight, such as branch 5 weight 0.7>0.4, retain branch 5, remove the branch with lower weight, remove branch 6; finally integrate all retained branch hypotheses, such as retaining branches 1, 2, and 5, and organize the trajectory point sequence in time order, such as the target 1 output sequence [t1 position 100 meters, t2 position 105 meters,...], the target 2 sequence [t1 position 110 meters, t2 position 115 meters,...] as a multi-target trajectory sequence. For example, in a target occlusion scene, 5 branches conflict, and after this resolution, 3 final trajectories are output.

[0112] In practical applications, in a flight target tracking scenario, to address the conflict caused by the intersection of multiple target trajectories, a multi-target trajectory decision is made based on the trajectory decoupling features output in 104 steps. For example, the feature vector of target A is [2.1, 3.0], that of target B is [1.8, 2.9], and that of target C is [-0.5, 4.2]. First, feature similarity is calculated to generate association weights: the dot product of the feature vectors of targets A and B is 2.1×1.8+3.0×2.9=3.78+8.7=12.48, which, after normalization, yields an association weight of 0.85; the dot product of targets A and C is 2.1×(-0.5)+3.0×4.2=-1.05+12.6=11.55, with a weight of 0.78. These weights are then input into a multi-hypothesis tracking algorithm to generate three initial trajectory branch hypotheses: branch 1 corresponds to target A, branch 2 corresponds to target B, and branch 3 corresponds to target C. Trajectory collisions are detected at a time step of 50 seconds: the real-time spatial positions of branch 1 and branch 2 are (2050m, 3100m) and (2045m, 3095m) respectively, and the spatial distance is... Less than the dynamic threshold of 10m. Calculate the instantaneous velocity vectors of the two branches: Branch 1 velocity. Branch 2 speed Vector angle Since the set angle is below 60°, two branches are retained; however, the angle between branch 1 and branch 3 is 78° (exceeding 60°) and the association weight is 0.85 > 0.78, so branch 1 is retained. After integration, three clear trajectory sequences are output, effectively resolving the tracking conflict problem in the intersection area.

[0113] In the overall scheme of step 105 above, correlation weights are generated by utilizing trajectory decoupling features and input into a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses. When facing trajectory intersection or occlusion scenarios, the instantaneous velocity vector angle is innovatively introduced as the core criterion for trajectory conflict discrimination. When the spatial distance between two branches is less than the dynamic adaptive threshold and the velocity vector angle exceeds the set angle threshold, the system automatically triggers a branch filtering mechanism based on correlation weights to accurately retain the trajectory branch with higher feature matching degree. The entire conflict resolution process deeply integrates the three discrimination criteria of spatial distance constraint, motion direction difference quantification, and feature similarity. The final output multi-target trajectory sequence has strong spatiotemporal continuity, trajectory ambiguity resolution capability, and dynamic scene adaptability, providing a robust solution for centimeter-level trajectory tracking of cluster targets in strong occlusion and high-density intersection environments.

[0114] The following is a complete embodiment for steps 101 to 105:

[0115] like Figure 2 As shown, in an air traffic monitoring scenario, the radar system tracks a group of flying targets that obstruct each other and whose trajectories intersect within the area. The system first continuously collects echo signals for 10 seconds at a pulse repetition frequency of 10 kHz, forming radar data corresponding to 100 pulses (10 kHz × 10 seconds = 100 pulses). Each pulse is processed to generate 200 distance sampling points, which are finally arranged in chronological order into a 100-row, 200-column two-dimensional radar echo matrix.

[0116] Signal separation is achieved by analyzing the characteristics of target scattering points. For example, a scattering point may exhibit linear characteristics by moving uniformly from distance unit 10 to 50 within 100 time steps, or it may exhibit surface characteristics by changing from position 15 to 48 along an arc-shaped path. For linearly distributed points, when target A is at position (35.0) units at step 50 and at position (35.2) units at step 51, satisfying a continuous and smooth change, they are classified as the same target. For surface-distributed points, when the curvature of the position change differs from that at step 25 to step 50 by less than 0.1 units and the position continuously increases throughout the process, they are determined to be the same target. This process ultimately outputs the three-dimensional positions of 100 scattering points, such as target A located at (30, 40, 50) units and its amplitude value of 3.2 units.

[0117] Based on the separated data, the target relationship is constructed. At a step size of 60, the spatial distance between target A (32,41,53) and target B (31,40,52) is calculated to be 1.73 units. Simultaneously, target A moves from (22,32,42) to (32,42,52) within 2 seconds at a calculated speed of approximately 8.66 units per second, while target B's speed is 7.8 units per second, with a speed difference of 0.86 units per second. A dynamic edge weight of 1.49 units squared per second is generated. When the distance between the two targets remains less than 0.5 units during a step size of 55 to 65, an occlusion relationship edge is established. The magnitude value of target A (3.2 units) is converted into a feature vector [1.5, -0.3]. Through weight matrix transformation, intermediate features [1.26, -0.03] are obtained. Then, neighboring node features, such as the weighted feature of node B [1.34, 0.60], are aggregated and concatenated, and the trajectory decoupling feature [0.19, 3.36, 0] is output after activation function.

[0118] Finally, a multi-target trajectory is generated. When the spatial distance between branch 1 (2105, 3150) and branch 2 (2100, 3145) is 7.07 meters and less than the 10-meter threshold, the velocity vectors of branch 1 (15, 20) and branch 2 (-5, 30) are retained simultaneously, and the angle between the two vectors is 46 degrees and less than the 60-degree threshold. When the angle between branch 1 and branch 3 exceeds the threshold of 78 degrees and the association weight is 0.85 and greater than 0.78, the branch with higher weight is retained first, and the decoupled trajectory sequence is finally output.

[0119] Figure 3 This application provides a schematic diagram of the structure of a radar signal multi-target tracking system based on a dynamic graph convolutional network, as shown in the embodiment of this application. Figure 3 As shown, the system includes:

[0120] Acquisition module 31 is used to acquire radar one-dimensional range image sequence of target cluster, perform pulse compression processing on the one-dimensional range image sequence, and arrange it according to the slow time dimension to form a radar echo two-dimensional sequence.

[0121] The separation module 32 is used to separate the position and amplitude of the aliased scattering points in the two-dimensional sequence of the radar echo according to the linear or curved distribution characteristics of the scattering points of the conical target, and obtain the scattering point position and amplitude data.

[0122] The construction module 33 is used to calculate the spatial distance between targets based on the position information in the scattering point position and amplitude data, and to construct dynamic edge weights by combining the radial velocity difference. When the spatial distance between targets is less than the radar resolution threshold, an occlusion relationship edge is generated.

[0123] The generation module 34 is used to initialize node feature weights based on the amplitude information in the scattering point position and amplitude data, use the dynamic edge weights as the neighborhood aggregation coefficients of the graph attention mechanism, aggregate the neighborhood node features of the occlusion relationship edges through the dynamic graph convolutional network, learn the decoupling features of the target trajectory in the occlusion and intersection regions, and generate trajectory decoupling features.

[0124] Output module 35 is used to generate association weights based on the trajectory decoupling features, input a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses, resolve conflicts based on the instantaneous velocity vector angle of the multi-branch trajectory hypotheses, and retain the branch trajectory hypothesis with higher association weight when the spatial distance between two branches is less than a dynamic threshold and the velocity vector angle exceeds a set angle, and output a multi-target trajectory sequence.

[0125] Figure 3 The aforementioned radar signal multi-target tracking system based on dynamic graph convolutional networks can perform... Figure 1 The implementation principle and technical effects of the radar signal multi-target tracking method based on dynamic graph convolutional networks described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit performs operations in the radar signal multi-target tracking system based on dynamic graph convolutional networks in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0126] In one possible design, Figure 3 The radar signal multi-target tracking system based on a dynamic graph convolutional network, as shown in the embodiment, can be implemented as a computing device, such as... Figure 4 As shown, the computing device may include a storage component 41 and a processing component 42;

[0127] The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 42.

[0128] The processing component 42 is used for the above Figure 1 The embodiment describes a multi-target radar signal tracking method based on dynamic graph convolutional networks.

[0129] The processing component 42 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0130] Storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0132] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0133] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0134] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0135] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The illustrated embodiment presents a radar signal multi-target tracking method based on a dynamic graph convolutional network.

[0136] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0137] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A radar signal multi-target tracking method based on a dynamic graph convolution network, characterized in that, The method comprises the following steps: Collecting a radar one-dimensional range image sequence of a target cluster, performing pulse compression processing on the one-dimensional range image sequence, and arranging the one-dimensional range image sequence in a slow time dimension to form a radar echo two-dimensional sequence; According to the linear or curved surface distribution characteristics of the scattering points of the cone target, the positions and amplitudes of the mixed scattering points in the radar echo two-dimensional sequence are separated to obtain scattering point position and amplitude data; Based on the position information in the scattering point position and amplitude data, the inter-target spatial distance is calculated, and a dynamic edge weight is constructed in combination with the radial velocity difference, and a shielding relationship edge is generated when the inter-target spatial distance is less than a radar resolution threshold; According to the amplitude information in the scattering point position and amplitude data, the node feature weight is initialized, the dynamic edge weight is taken as the neighborhood aggregation coefficient of the graph attention mechanism, the neighborhood node features of the shielding relationship edge are aggregated through the dynamic graph convolution network, the decoupling features of the target trajectories in the shielding and crossing regions are learned, and the trajectory decoupling features are generated; Based on the trajectory decoupling features, an association weight is generated, a multi-hypothesis tracking algorithm is input to generate multi-branch trajectory hypotheses, conflict resolution is performed based on the instantaneous speed vector angle of the multi-branch trajectory hypotheses, when the spatial distance of two branches is less than a dynamic threshold and the speed vector angle exceeds a set angle, the trajectory hypothesis with a higher association weight is retained, and a multi-target trajectory sequence is output.

2. The method of claim 1, wherein, According to the amplitude information in the scattering point position and amplitude data, the node feature weight is initialized, the dynamic edge weight is taken as the neighborhood aggregation coefficient of the graph attention mechanism, the neighborhood node features of the shielding relationship edge are aggregated through the dynamic graph convolution network, the decoupling features of the target trajectories in the shielding and crossing regions are learned, and the trajectory decoupling features are generated, comprising: Based on the amplitude information in the scattering point position and amplitude data, the amplitude value is converted into an initial node feature vector, and the node feature weight is initialized based on the initial node feature vector; The dynamic edge weight is assigned to the neighborhood aggregation coefficient of the graph attention mechanism, a linear transformation operation is performed on the initial node feature vector through a learnable weight matrix to generate an intermediate node feature; Through the dynamic graph convolution network, the shielding relationship edge is taken as a connection path, the intermediate node features of the neighborhood of the shielding relationship edge are weighted and summed to aggregate the neighborhood node features; The neighborhood node features after weighted summation and the intermediate node features of the current node are spliced, a nonlinear activation operation is performed on the spliced features, the decoupling features of the target trajectories in the shielding and crossing regions are learned, and an updated feature vector is output as the trajectory decoupling features.

3. The method of claim 1, wherein, Based on the position information in the scattering point position and amplitude data, the inter-target spatial distance is calculated, and a dynamic edge weight is constructed in combination with the radial velocity difference, and a shielding relationship edge is generated when the inter-target spatial distance is less than a radar resolution threshold, comprising: Based on the position information in the scattering point position and amplitude data, the inter-target spatial distance between any two scattering points is calculated, and the radial velocity difference of the scattering points is calculated according to the change of the position information in the slow time dimension; The inter-target spatial distance and the radial velocity difference are multiplied, and a dynamic edge weight is constructed based on the multiplication result. When the target-to-target distance is less than a radar resolution threshold, an occlusion relationship edge is generated between the corresponding scattering points, and the dynamic edge weight is used as the correlation strength value of the occlusion relationship edge.

4. The method of claim 1, wherein, Based on the trajectory decoupling features, correlation weights are generated, multi-hypothesis tracking algorithms are input to generate multi-branch trajectory hypotheses, and conflict resolution is performed based on the instantaneous speed vector angles of the multi-branch trajectory hypotheses. When the spatial distance of two branches is less than a dynamic threshold and the speed vector angle exceeds a set angle, the branch trajectory hypothesis with a higher correlation weight is retained, and a multi-target trajectory sequence is output, including: Based on the trajectory decoupling features, feature vector similarities of different trajectory points are calculated, correlation weights are generated, and the correlation weights are input to a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses. For the multi-branch trajectory hypotheses, when the spatial distance of any two branches is less than a dynamic threshold, the instantaneous speed vector angle of the two branches is calculated. When the instantaneous speed vector angle exceeds a set angle, the correlation weights of the two branches are compared, the branch trajectory hypothesis with a higher correlation weight is retained, and the retained branch trajectory hypotheses are integrated and output in time sequence as a multi-target trajectory sequence.

5. The method of claim 4, wherein, For the multi-branch trajectory hypotheses, when the spatial distance of any two branches is less than a dynamic threshold, the instantaneous speed vector angle of the two branches is calculated. When the instantaneous speed vector angle exceeds a set angle, the correlation weights of the two branches are compared, the branch trajectory hypothesis with a higher correlation weight is retained, and the retained branch trajectory hypotheses are integrated and output in time sequence as a multi-target trajectory sequence, including: The multi-branch trajectory hypotheses with a spatial distance less than a dynamic threshold at the current time are screened to form a set of branch pairs to be processed, and the instantaneous speed vectors of each branch trajectory pair at the current time are extracted from the set of branch pairs to be processed. The angle value of the two instantaneous speed vectors in the same branch trajectory pair is calculated, and the angle value is obtained by vector dot product operation and arccosine function conversion. When the angle value exceeds a set angle, the correlation weight values of the two branch trajectories are extracted from the set of branch pairs to be processed. The correlation weight values of the two branch trajectories are compared, the branch trajectory with a larger correlation weight value is retained, and the branch trajectory with a smaller correlation weight value is removed. The retained branch trajectory hypotheses are integrated and output in time sequence as a multi-target trajectory sequence.

6. The method of claim 1, wherein, According to the linear or curved surface distribution characteristics of the scattering points of the cone target, the positions and amplitudes of the mixed scattering points in the radar echo two-dimensional sequence are separated to obtain scattering point position and amplitude data, including: The position information of each scattering point in the radar echo two-dimensional sequence is extracted in the slow time dimension, the continuous shift pattern of the sequence is analyzed, and the position distribution characteristics of the sequence are identified as linear distribution characteristics or curved surface distribution characteristics. For the linear distribution characteristics of the scattering points of the cone target, the scattering points that meet the continuity condition are divided into the same target according to the continuity of the position shift direction of adjacent slow time frames and the smoothness of the amplitude change. The curvature change value of the scattering point position change sequence is calculated according to the curved surface distribution characteristics of the scattering points of the cone target, and when the curvature change value is less than a preset threshold value and the position offset amplitude remains continuously increasing, it is determined that the scattering points belong to the same target; Based on the scattering points divided into the same target, the spatial position overlapping position information is detected, the overlapping scattering point positions and amplitudes are separated, and separated scattering point position and amplitude data are obtained.

7. The method of claim 2, wherein, Through a dynamic graph convolution network, the intermediate node features of the neighborhood of the occlusion relationship edge are aggregated by weighted summation, including: Based on the occlusion relationship edge, all neighborhood nodes directly connected to the current node through the occlusion relationship edge are screened out to form a neighborhood node set; The intermediate node features of each node in the neighborhood node set are extracted, and the intermediate node features are generated by linear transformation from the initial node features; The dynamic edge weight is taken as the neighborhood aggregation coefficient, and the intermediate node features of each neighborhood node are multiplied by the corresponding neighborhood aggregation coefficient to generate weighted neighborhood node features; The weighted neighborhood node features of the same current node are summed to output the aggregated neighborhood node features.

8. A radar signal multi-target tracking system based on dynamic graph convolutional network, characterized in that, It includes: The acquisition module is used for acquiring a radar one-dimensional range image sequence of a target cluster, performing pulse compression processing on the one-dimensional range image sequence, and arranging the one-dimensional range image sequence in the slow time dimension to form a radar echo two-dimensional sequence; The separation module is used for separating the positions and amplitudes of the overlapping scattering points in the radar echo two-dimensional sequence according to the linear or curved surface distribution characteristics of the scattering points of the cone target, and obtaining scattering point position and amplitude data; The construction module is used for calculating the inter-target spatial distance based on the position information in the scattering point position and amplitude data, and constructing a dynamic edge weight in combination with a radial velocity difference, and generating an occlusion relationship edge when the inter-target spatial distance is less than a radar resolution threshold value; The generation module is used for initializing node feature weights according to the amplitude information in the scattering point position and amplitude data, taking the dynamic edge weight as the neighborhood aggregation coefficient of the graph attention mechanism, aggregating the neighborhood node features of the occlusion relationship edge through a dynamic graph convolution network, learning the decoupling features of the target trajectories in the occlusion and intersection regions, and generating trajectory decoupling features; The output module is used for generating correlation weights based on the trajectory decoupling features, inputting a multi-hypothesis tracking algorithm to generate multi-branch trajectory hypotheses, performing conflict resolution based on the instantaneous velocity vector angles of the multi-branch trajectory hypotheses, retaining the branch trajectory hypothesis with a higher correlation weight when the spatial distance of two branches is less than a dynamic threshold value and the velocity vector angle exceeds a set angle, and outputting a multi-target trajectory sequence.

9. A computing device, comprising: The processing component and the storage component are included; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the radar signal multi-target tracking method based on the dynamic graph convolution network in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer program is stored, and when the computer program is executed by a computer, the radar signal multi-target tracking method based on the dynamic graph convolution network in any one of claims 1 to 7 is realized.

Images