A method and system for detecting and tracking a wideband radar-based unmanned aerial vehicle cluster extended target
By treating the UAV swarm as multiple scattering centers, combining noise and interference modeling, and employing a multi-beam coherent fitting method, the problems of multi-point tracking and false alarms in UAV swarm target detection and tracking by broadband radar were solved, achieving stable target detection and tracking, and improving environmental adaptability and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF SCI & TECH
- Filing Date
- 2026-03-25
- Publication Date
- 2026-07-14
AI Technical Summary
Existing broadband radars face problems such as multi-point tracking, multi-peak response, false alarms, trajectory drift, and unstable observation in UAV swarm target detection and tracking, making it difficult to achieve stable continuous output and resist false correlation in complex environments.
By equating the UAV swarm to an extended target with multiple scattering centers, combining noise and interference modeling to generate radar echoes, using a multi-beam coherent fitting method for angle estimation, and incorporating the spatial shape parameters of the swarm target as state variables into the tracking model, a frame-by-frame closed-loop iterative update is performed to achieve dynamic control of the candidate observation range.
It improves the continuous output capability and anti-false association performance in complex environments, and enhances the environmental adaptability and stability of target detection and tracking.
Smart Images
Figure CN121918110B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar detection and target tracking technology, and in particular to a method and system for extended target detection and tracking of UAV swarms based on broadband radar. Background Technology
[0002] With the rapid popularization of drones in civilian and industrial scenarios, the formation flight and collaborative operation modes of multiple drones are gradually increasing. Compared with single targets, drone swarm targets are characterized by large numbers, small scale, weak scattering, high maneuverability and variable formation. In complex electromagnetic and clutter environments, they are more prone to low signal-to-noise ratio, significant false alarm interference, and increased observation fluctuations, making traditional radar detection and tracking methods based on the core assumption of single targets difficult to apply directly.
[0003] Wideband radar is widely used for small target detection due to its high range resolution. Current technologies generally follow a mature approach of data association and recursive estimation based on observation data, and various association and filtering frameworks have been developed to improve trajectory maintenance capabilities under false alarm and missed detection conditions. Meanwhile, some research has begun to focus on the overall characterization and tracking of formations and groups of targets. However, when observing UAV formations / swarms with wideband radar, high resolution often leads to multiple traces and multi-peak responses in a single frame with significant fluctuations in the number of traces. This is easily intertwined with false alarms caused by clutter and noise, making the traditional "single-point observation" assumption more prone to failure, leading to false associations, trajectory drift, or discontinuity. Simply compressing multiple traces into a single observation to reduce complexity may weaken the stable characterization of the entire swarm, causing output jitter and making it more sensitive to formation changes and short-term missing measurements. Furthermore, many schemes still rely heavily on fixed thresholds or empirical rules to constrain the candidate range for the next frame, lacking a mechanism for adaptive closed-loop constraints using historical estimation results. This results in insufficient re-acquisition capability, limited continuity, and limited robustness under conditions of increased interference or missing observations.
[0004] In the theory of radar echo modeling and observational characterization, existing research is mostly based on statistical fluctuation models and the assumption of idealized scatterers. Targets are treated as isolated point targets or sets of weakly coupled scattering centers to construct echo and measurement models. This approach can complete echo simulation and measurement generation at a relatively low computational cost and is easily integrated with classical detection, correlation, and recursive estimation frameworks. However, when dealing with group targets such as UAV formations and swarms, the relative geometric relationships and formation evolution between targets can cause multi-peak responses, point fragmentation / aggregation, and fluctuations in the number of observations. This makes the modeling and processing chain, which relies on the assumption of static scattering or single-point observation, more prone to overall instability and error accumulation, affecting the reliability of frame-by-frame output. To address these issues, existing research has made some progress in areas such as overall characterization of group targets, point merging and robust correlation, and candidate range constraints based on prior information. However, under the high-resolution observation conditions of broadband digital array radar, how to form a more representative overall observation that conforms to the observation characteristics of extended group targets and serves stable closed-loop tracking without significantly increasing computational complexity remains to be further explored.
[0005] Therefore, there is an urgent need for a holistic detection and tracking method for broadband radar UAV swarm targets, which can construct robust holistic representative observations under multi-point trace conditions and achieve stable output through frame-by-frame closed-loop iteration, thereby improving the continuous output capability and anti-false correlation performance in complex environments. Summary of the Invention
[0006] The purpose of this invention is to provide a broadband radar-based UAV swarm extended target detection and tracking method and system with a concise model, high descriptive efficiency, and high target tracking accuracy, thereby improving continuous output capability and anti-false association performance in complex environments, and enhancing the environmental adaptability and stability of target detection and tracking.
[0007] The technical solution to achieve the purpose of this invention is: a method for extended target detection and tracking of UAV swarms based on broadband radar, comprising the following steps:
[0008] Step 1: Set up a drone swarm scenario, calculate the target electromagnetic scattering characteristics based on the scenario, and treat the drone swarm as an extended target composed of multiple scattering centers. Combine noise and interference modeling to generate broadband radar echoes of the extended target of the swarm.
[0009] Step 2: Calibrate and compensate the broadband observation data. If no candidate target is found, wait for the next frame of data input. When the target distance information is extracted in the distance dimension, use the multibeam coherent fitting method to estimate the angle and jointly output the cluster target distance-angle detection result.
[0010] Step 3: Perform data association based on joint information, introduce the spatial shape parameters of the cluster targets as state variables into the tracking model, and predict and update them together with the target motion state to achieve dynamic control of the candidate observation range and improve the continuity and robustness of subsequent tracking.
[0011] A broadband radar-based extended target detection and tracking system for UAV swarms, the system being used to implement the aforementioned broadband radar-based extended target detection and tracking method for UAV swarms, the system comprising:
[0012] The broadband radar modeling module for cluster targets sets up a UAV cluster scenario, calculates the electromagnetic scattering characteristics of the target based on the scenario, and equates the UAV cluster to an extended target composed of multiple scattering centers. It then combines noise and interference modeling to generate broadband radar echoes of the cluster extended target.
[0013] The cluster target broadband radar detection module calibrates and compensates the broadband observation data. If no candidate target is found, it waits for the next frame of data input. When the target distance information is extracted in the distance dimension, the multi-beam coherent fitting method is used to estimate the angle, and the cluster target distance-angle detection result is jointly output.
[0014] The cluster target broadband radar tracking module performs data association based on joint information, introduces the spatial shape parameters of the cluster targets as state variables into the tracking model, and predicts and updates them together with the target motion state, thereby realizing dynamic control of the candidate observation range and improving the continuity and robustness of subsequent tracking.
[0015] A mobile terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned broadband radar-based UAV swarm extended target detection and tracking method.
[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the broadband radar-based UAV swarm extended target detection and tracking method.
[0017] Compared with the prior art, the significant advantages of this invention are:
[0018] (1) Based on the high-resolution observation characteristics of broadband radar for UAV formations and clusters, a cluster extended target echo model is constructed based on the electromagnetic scattering characteristics of the target. The UAV cluster is equivalent to an overall extended target composed of multiple scattering centers for detection and modeling. This allows for a more realistic reflection of the spatial structure and energy distribution characteristics of the cluster target under broadband observation conditions, providing a reliable observation basis for subsequent stable detection and cross-frame correlation.
[0019] (2) For the detected candidate range gates, the coherent observation information of multiple scanning beams in the neighborhood of the main lobe is used to make precise angle estimation. This avoids the dependence of traditional single-pulse angle measurement methods on linear assumptions and symmetric structures, effectively suppresses systematic angle measurement deviations under beam edge and extended target conditions, and improves the consistency and robustness of angle estimation.
[0020] (3) In the target tracking stage, the spatial shape parameters of the cluster targets are introduced into the state model as predictable state variables. Through the frame-by-frame closed-loop mechanism, the historical estimation and prediction prior feedback are used for the candidate range constraint and gating adaptive control of the next frame, realizing the dynamic optimization of the observation search area. Thus, the trajectory continuity and stable output can still be maintained under the conditions of short-term missing measurement, enhanced interference and changes in cluster shape, improving the continuous output capability and anti-false association performance in complex environments, and improving the environmental adaptability and stability of target detection and tracking. Attached Figure Description
[0021] Figure 1 This is a flowchart illustrating the extended target detection and tracking method for UAV swarms based on broadband radar according to the present invention.
[0022] Figure 2 This is a comparison curve of the Rayleigh clutter modeling probability distribution value and the theoretical value in an embodiment of the present invention.
[0023] Figure 3 This is a graph showing the energy superposition distance migration curves for each frame in an embodiment of the present invention.
[0024] Figure 4 This is a graph showing the superposition of energy values for each frame after resampling in an embodiment of the present invention.
[0025] Figure 5 This is a comparison chart of the azimuth measurement points and ideal points of each frame of the UAV formation in this embodiment of the invention.
[0026] Figure 6 This is a comparison chart of the measured pitch angle points of each UAV formation and the ideal points in each frame of this invention embodiment.
[0027] Figure 7 This is a comparison chart of the three-dimensional measurement points and ideal points of each frame of the UAV formation in this embodiment of the invention.
[0028] Figure 8 These are the three-dimensional measurement points and overall predicted points of the UAV formation in each frame of this embodiment of the invention. Detailed Implementation
[0029] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0030] like Figure 1As shown, this invention provides an extended target detection and tracking method for UAV swarms based on broadband radar, comprising the following steps:
[0031] Step 1: Set up a drone swarm scenario, calculate the target electromagnetic scattering characteristics based on the scenario, and treat the drone swarm as an extended target composed of multiple scattering centers. Combine noise and interference modeling to generate broadband radar echoes of the extended target of the swarm.
[0032] Step 2: Calibrate and compensate the broadband observation data. If no candidate target is found, wait for the next frame of data input. When the target distance information is extracted in the distance dimension, use the multibeam coherent fitting method to estimate the angle and jointly output the cluster target distance-angle detection result.
[0033] Step 3: Perform data association based on joint information, introduce the spatial shape parameters of the cluster targets as state variables into the tracking model, and predict and update them together with the target motion state to achieve dynamic control of the candidate observation range and improve the continuity and robustness of subsequent tracking.
[0034] As a specific example, step 1 involves setting up a drone swarm scenario, calculating the target's electromagnetic scattering characteristics based on the scenario, and equating the drone swarm to an extended target composed of multiple scattering centers. This is then combined with noise and interference modeling to generate a broadband radar echo of the extended target swarm, as detailed below:
[0035] Step 1: Set up a drone swarm scenario, calculate the target's electromagnetic scattering characteristics based on the scenario, and treat the drone swarm as an extended target composed of multiple scattering centers. Combine noise and interference modeling to generate broadband radar echoes of the extended target swarm, as detailed below:
[0036] Step 1.1: Set up a drone swarm scenario. Based on the geometric shape, attitude, and observation direction relationship of the drone swarm target, the drone array target is equivalent to an extended target composed of multiple scattering centers. Obtain the scattering center information of the swarm target, including equivalent scattering position parameters and scattering intensity parameters. Superimpose the echoes from each scattering center to obtain the interference-free swarm target received echo. The expression is:
[0037]
[0038] in, The received interference-free echo signal, The number of scattering centers For wavelength, For the first Radar cross-section (RCS) of each scattering center For the first The distance between each scattering center and the radar. For radar in the Gain in the direction of each scattering center and They are the first The azimuth and elevation angles of each scattering center relative to the radar. It is a rectangular pulse function. To save time, The pulse width. The rate of change of frequency, For carrier frequency, For the first Propagation delay at each scattering center The imaginary unit, It is an exponential function;
[0039] Step 1.2: Model the uncertainties in the observation background. These uncertainties include system noise and environmental clutter. The uncertainties are incorporated into the broadband radar echoes of the cluster targets to form echo data simulating the actual scene. The environmental clutter follows a Rayleigh distribution, expressed as:
[0040]
[0041] in, For clutter signals that conform to a Rayleigh distribution, It is the clutter intensity coefficient. It is the filter impulse response used to construct the power spectrum; They are the in-phase components of mutually independent zero-mean, unit-variance Gaussian white noise processes. They are the orthogonal components of mutually independent zero-mean, unit-variance Gaussian white noise processes; For filter energy, It is the integral variable.
[0042] As a specific example, in step 2, the broadband observation data is calibrated and compensated. If no candidate target is found, the system waits for the next frame of data input. When target distance information is extracted in the distance dimension, a multi-beam coherent fitting method is used for angle estimation, and the cluster target distance-angle detection result is jointly output, as follows:
[0043] Step 2.1: Due to its high range resolution, broadband radar causes significant changes in round-trip propagation delay over slow time due to the radial motion of the target within the observation window. This results in cross-cell migration of target energy in the range dimension, known as range migration. In broadband systems, this effect can be equivalent to the coupling between range frequency domain variables and slow time, causing non-stationary drift of the range image in slow time. Range migration severely impacts coherent accumulation, affecting target detection. To suppress the impact of range migration on subsequent range peak stability and angle estimation consistency, broadband echo data undergoes slow-time dimension resampling processing, expressed as:
[0044]
[0045] in, This is the broadband observation data after resampling correction. This is the original array multi-channel broadband observation data. For observation data in the fast time-frequency domain; For slow time, Indicates about fast time Fourier transform, Indicates frequency with fast time Inverse Fourier transform, The slow time coordinates are after resampling;
[0046] Step 2.2 Compared to Broadband observation data is aligned in the range dimension, which keeps the range response of the same target concentrated in the slow time series, thus providing a consistent input for stable range detection. Since UAV swarm targets are equivalent to extended targets, their range profiles often exhibit a multi-peak structure or local clustered strong scattering. First, the data is statistically accumulated in the range domain, and then an adaptive threshold is calculated based on the accumulated value. The threshold is used to extract peak values, so as to balance the detectability of weak targets and the distinguishability of multiple targets.
[0047] Broadband observation data after resampling correction Above, a distance dimension statistic is formed. The expression is:
[0048]
[0049] in, For slow-time sets, The magnitude of the complex amplitude;
[0050] By performing incoherent accumulation on slow-time sets, a distance criterion that is more sensitive to the peak value in the distance dimension can be obtained; an adaptive threshold in the distance dimension is introduced. right Local peak extraction is performed to obtain a set of candidate distance gates. And the corresponding distance, expressed as:
[0051]
[0052] in, For the first One candidate distance gate; This is a local peak operator used to extract representative distance peaks from a continuous region that passes through a threshold; For the first Distance estimation for each candidate distance gate; The speed of light;
[0053] If the candidate distance gate set is not empty, calculate the target distance corresponding to the peak value; otherwise, determine that there is no target in this frame and wait for the next frame's observation input.
[0054] Step 2.3: For candidate distance gates ,exist The array observation vector is extracted and a angular domain response is formed on a preset scanning angular grid to determine the coarse-angle unit corresponding to the main response peak, which serves as the initial estimate of the target angle. Within the main lobe neighborhood corresponding to the coarse-angle unit, multiple adjacent scanning beam directions are selected, and the scanning beams at the candidate range gate are extracted. The complex observation output at the specified location is coherently accumulated in the slow time dimension to construct a multibeam coherent observation vector:
[0055]
[0056] in, Indicates the first Frames in candidate distance gate Place, No. Coherent observations along the scanning beam direction; , The number of adjacent scan beams selected; Indicates the first Frames in candidate distance gate Multibeam coherent observation vector at location; superscript It is the transpose symbol;
[0057] Based on the geometric parameters of the radar array, the target's position at a given angle is established. The theoretical response vector of the multibeams under the following conditions for:
[0058]
[0059] in, These are the target azimuth and elevation angles, respectively. For the first The theoretical response value in each scanning beam direction, ;
[0060] No. The theoretical response in each scanning beam direction can be expressed as a coherent superposition of the array direction responses as follows:
[0061]
[0062] in, , These represent the number of array elements in the horizontal and vertical directions, respectively. These are the array element numbers for the horizontal and vertical directions, respectively. For the array element space coordinates, For the first Each scanning beam array element The corresponding weighting coefficients;
[0063] Define the multibeam coherence matching metric function as follows:
[0064]
[0065] in, The value of the multibeam coherent matching metric function. for The conjugate transpose of;
[0066] The matching metric function eliminates the influence of differences in gain between different scanning beams and echo amplitude fluctuations on the angle measurement results through energy normalization; a local search is performed within the coarse angle cell corresponding to the main response peak to determine the angle parameter that maximizes the matching metric function. This refers to the estimated results of the azimuth and elevation angles;
[0067] Step 2.4, for the first Candidate distance gate for frames The obtained distance and angle estimates are combined to obtain a joint measurement vector:
[0068]
[0069] in, For the first Frame number The joint measurement vector of the candidate distance gates, , , The first Frame number Distance estimation, azimuth estimation, and pitch estimation for each candidate distance gate;
[0070] To improve the availability and robustness of the joint output, for each joint measurement A reasonableness assessment is conducted, based on pre-defined constraints and consistency requirements, to determine whether the joint measurement meets physical feasibility, observation consistency, and output reliability; when the joint measurement... If the reasonableness condition is not met, the joint measurement is deemed unreliable and removed; if the reasonableness condition is met, the joint measurement is retained and added to the output set. .
[0071] As a specific example, step 3 performs data association based on joint information, introducing the spatial shape parameters of the cluster targets as state variables into the tracking model. These parameters are predicted and updated together with the target motion state. Shape prediction is used to construct shape-aware gating conditions and generate an adaptive search window. This enables dynamic control of the candidate observation range, suppression of erroneous associations, and enhanced recapture in scenarios with missing data during frame-by-frame processing. Consequently, the cluster target tracking results are output, and the data processing is updated based on the tracking results to improve the continuity and robustness of subsequent tracking. The details are as follows:
[0072] Step 3.1, place the first The set of joint measurements retained after the frame passes the rationality judgment is denoted as . To suppress the multi-peaked traces and local false alarms generated by broadband extended targets within the same frame, the following measures are taken: Intra-frame point merging is performed, defining intra-frame nearest neighbor criteria based on three dimensions: distance, azimuth, and pitch. Measurements satisfying the nearest neighbor relationship are grouped into the same cluster, and the intra-frame candidate point set of this cluster is: , Indicates the first Frame number The number of merged points within a cluster is used to generate the corresponding number of points in the cluster statistical representative value. Frame number Fusion measurement of individual clusters ;
[0073] Will Translated to Cartesian coordinates, the three-dimensional Cartesian uniform state and cluster shape measurement matrix are defined as follows:
[0074]
[0075]
[0076] in, For state estimation, The state matrix; For the first Frame number A shape measurement matrix for each cluster, used to characterize the spatial scale and shape of the cluster; subscript Indicates frame number and index Indicates the cluster sequence number; Indicates the first Frame number Within the cluster Cartesian coordinate vectors of points; It is a positive number less than a set value, used to prevent the matrix from being non-positive definite; It is a three-dimensional identity matrix;
[0077] To achieve stable cross-frame correlation, the first frame... Shape-aware gating is set for the predictive prior of each cluster:
[0078]
[0079] in, For the first Frame number The predicted state value of each cluster. For the first Frame number The state prediction covariance matrix of each cluster The noise covariance is measured as a basis. For the first The shape prediction matrix of each cluster The number of valid points is estimated based on data from neighboring frames. This is the gate threshold;
[0080] Step 3.2: After completing the cross-frame point association, in order to smoothly and robustly track the cross-frame changes of the fused measurement points, a filtering recursion is performed on the 3D Cartesian uniform velocity model. Adaptive measurement noise with point count and shape distribution is introduced, so that the filtering gain and gating strength are automatically adjusted with the cluster scale. The expression for the state update process is:
[0081]
[0082]
[0083]
[0084]
[0085]
[0086]
[0087] in For the first Frame number The adaptive measurement noise covariance of each cluster enables the filter gain to be automatically adjusted as the number of points and their distribution change, thereby improving stability and reducing the risk of sudden error changes in multi-target scenarios. For the innovation covariance in the update phase, For the first The state filter estimate after frame update For the first Covariance of the state filter estimate after frame update; The filter gain is used for an adaptive trade-off between the predicted prior and the current observation, when When the variance is large or the prior variance of the prediction is small, the update depends more on the prediction; otherwise, the update depends more on the observation, thus achieving smooth tracking of the fused measurements. This is the shape smoothing coefficient, used to weight the shape prediction and shape measurement. The identity matrix is consistent with the dimension of the state;
[0088] Step 3.3: To perform gating association and maintain trajectory continuity in the next frame, based on the inter-frame time interval... The state is predicted, and the shape matrix is predicted by random walk and positive definite constraint processing is applied.
[0089] The expressions for the state transition matrix and the prediction process are as follows:
[0090]
[0091]
[0092]
[0093] in, This is the state transition matrix; For the next frame, the predicted filter estimate, The covariance estimated by the prediction filter for the next frame. For the first The shape prediction matrix for the next frame of each cluster. The noise covariance of the state process. For shape process noise covariance, This involves symmetricizing the matrix and imposing a lower bound constraint on the eigenvalues to ensure that the result is a symmetric positive definite matrix. Indicates the next frame prediction fusion measurement;
[0094] Step 3.4: To reduce the search range in the next frame and make the search window adapt to the cluster scale, construct the predicted measurement scatter based on shape prediction and generate a spherical coordinate domain window:
[0095]
[0096] in, For the first Frame number The search window vector for each cluster, The width of the search window is the distance dimension. The width of the azimuth search window. The width of the search window is the pitch angle dimension.
[0097] As a specific example, in step 3.1, for each fused measurement in the current frame, if the fused measurement passes the prediction gate of any existing trajectory, then the fused measurement is associated with the corresponding trajectory; if the fused measurement does not pass the prediction gate of any existing trajectory, then the fused measurement is used as the initial measurement for the new trajectory; for trajectories that are not associated with any measurement in multiple consecutive frames, they are maintained or terminated according to a preset strategy.
[0098] As a specific example, in step 3, to achieve cyclical data updates, the prediction fusion measurement is... The gating center is used for step 3.1 in the next frame. At the same time, the shape prediction matrix and prediction covariance adaptively determine the scale and direction of the gating, so as to automatically widen the gating range when the cluster scale expands and automatically tighten the gating range when the cluster scale shrinks. When the trajectory is stable, the gating threshold is appropriately tightened to reduce false alarms and false associations; when the trajectory shows maneuvering or short-term missing detection, the gating threshold is appropriately widened to improve the re-acquisition probability.
[0099] As a specific example, step 3 further forms a continuous closed-loop processing chain:
[0100] At the beginning of each frame processing, the estimation results of the previous frame are used to generate the prior prediction of the next frame, adaptively limiting the candidate observation range and completing the gating screening; cross-frame association is performed on the candidate observations that pass the gating to form the effective overall observation input of the current processing frame, and the overall state and shape of the cluster are recursively updated; the updated state is then used to generate the prediction prior of the next frame and fed back to the subsequent gating and association links, realizing frame-by-frame output and improving the continuity and robustness of subsequent tracking.
[0101] This invention also provides a broadband radar-based extended target detection and tracking system for UAV swarms. This system is used to implement the aforementioned broadband radar-based extended target detection and tracking method for UAV swarms. The system includes:
[0102] The broadband radar modeling module for cluster targets sets up a UAV cluster scenario, calculates the electromagnetic scattering characteristics of the target based on the scenario, and equates the UAV cluster to an extended target composed of multiple scattering centers. It then combines noise and interference modeling to generate broadband radar echoes of the cluster extended target.
[0103] The cluster target broadband radar detection module calibrates and compensates the broadband observation data. If no candidate target is found, it waits for the next frame of data input. When the target distance information is extracted in the distance dimension, the multi-beam coherent fitting method is used to estimate the angle, and the cluster target distance-angle detection result is jointly output.
[0104] The cluster target broadband radar tracking module performs data association based on joint information, introduces the spatial shape parameters of the cluster targets as state variables into the tracking model, and predicts and updates them together with the target motion state, thereby realizing dynamic control of the candidate observation range and improving the continuity and robustness of subsequent tracking.
[0105] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned method for extended target detection and tracking of UAV swarms based on broadband radar.
[0106] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the described method for extended target detection and tracking of UAV swarms based on broadband radar.
[0107] Example
[0108] like Figure 1 As shown, the UAV swarm extended target detection and tracking method based on broadband radar provided in this embodiment includes the following steps:
[0109] Step 1: Set up a drone swarm scenario, calculate the target electromagnetic scattering characteristics based on the scenario, and treat the drone swarm as an extended target composed of multiple scattering centers. Combine noise and interference modeling to generate broadband radar echoes of the extended target of the swarm.
[0110] Step 2: Calibrate and compensate the broadband observation data. If no candidate target is found, wait for the next frame of data input. When the target distance information is extracted in the distance dimension, use the multibeam coherent fitting method to estimate the angle and jointly output the cluster target distance-angle detection result.
[0111] Step 3: Perform data association based on joint information, introduce the spatial shape parameters of the cluster targets as state variables into the tracking model, and predict and update them together with the target motion state to achieve dynamic control of the candidate observation range and improve the continuity and robustness of subsequent tracking.
[0112] The broadband radar parameters used in the simulation are as follows:
[0113] The carrier frequency is 5 GHz, the pulse repetition frequency is 1000, the bandwidth is 400 MHz, the number of horizontal array elements is 32, and the number of vertical array elements is 32.
[0114] Figure 2 The comparison between the Rayleigh clutter modeling probability distribution value and the theoretical value demonstrates the accuracy of the clutter modeling. Figure 3This is a schematic diagram of the range migration caused by the superposition of pulse energies within a frame. It can be seen that, without compensation, the peak values of each extended target broaden in the range dimension. Figure 4 This is a schematic diagram of the superposition of pulse energies within a frame after resampling. The superimposed extended target no longer expands in the range dimension. Figure 5 It compares the azimuth measurement point mark with the ideal point mark. Figure 6 It compares the pitch angle measurement point with the ideal point. Figure 7 By comparing the three-dimensional measurement point traces with the ideal point traces, it was shown that the three-dimensional measurement point traces and the ideal point traces have a high degree of matching, which verifies the accuracy of the distance and angle measurement method of the present invention. Figure 8 The paper further demonstrates the comparison between the fused measurement prediction points and the measured points of the cluster starting from the second frame, verifying the accuracy of the method of the present invention in calculating the predicted points.
[0115] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for extended target detection and tracking of UAV swarms based on broadband radar, characterized in that, Includes the following steps: Step 1: Set up a drone swarm scenario, calculate the target electromagnetic scattering characteristics based on the scenario, and treat the drone swarm as an extended target composed of multiple scattering centers. Combine noise and interference modeling to generate broadband radar echoes of the extended target of the swarm. Step 2: Calibrate and compensate the broadband observation data. If no candidate target is found, wait for the next frame of data input. When the target distance information is extracted in the distance dimension, use the multibeam coherent fitting method to estimate the angle and jointly output the cluster target distance-angle detection result. Step 3: Perform data association based on joint information, introduce the spatial shape parameters of the cluster targets as state variables into the tracking model, and predict and update them together with the target motion state to achieve dynamic control of the candidate observation range and improve the continuity and robustness of subsequent tracking. Step 1 is as follows: Step 1.1: Set up a drone swarm scenario. Based on the geometric shape, attitude, and observation direction relationship of the drone swarm target, the drone array target is equivalent to an extended target composed of multiple scattering centers. Obtain the scattering center information of the swarm target, including equivalent scattering position parameters and scattering intensity parameters. Superimpose the echoes from each scattering center to obtain the interference-free swarm target received echo. The expression is: in, The received interference-free echo signal, The number of scattering centers For wavelength, For the first Radar cross-section (RCS) of each scattering center For the first The distance between each scattering center and the radar. For radar in the Gain in the direction of each scattering center and They are the first The azimuth and elevation angles of each scattering center relative to the radar. It is a rectangular pulse function. To save time, The pulse width. The rate of change of frequency, For carrier frequency, For the first Propagation delay at each scattering center The imaginary unit, It is an exponential function; Step 1.2: Model the uncertainties in the observation background. These uncertainties include system noise and environmental clutter. The uncertainties are incorporated into the broadband radar echoes of the cluster targets to form echo data simulating the actual scene. The environmental clutter follows a Rayleigh distribution, expressed as: in, For clutter signals that conform to a Rayleigh distribution, It is the clutter intensity coefficient. It is the filter impulse response used to construct the power spectrum; They are the in-phase components of mutually independent zero-mean, unit-variance Gaussian white noise processes. They are the orthogonal components of mutually independent zero-mean, unit-variance Gaussian white noise processes; For filter energy, It is the integral variable.
2. The method for extended target detection and tracking of UAV swarms based on broadband radar according to claim 1, characterized in that, Step 2 involves calibrating and compensating the broadband observation data. If no candidate target is found, the system waits for the next frame of data input. When target distance information is extracted in the distance dimension, a multi-beam coherent fitting method is used for angle estimation, and the cluster target distance-angle detection results are jointly output, as follows: Step 2.1: To suppress the impact of range migration on the stability of subsequent range peak values and the consistency of angle estimation, the broadband echo data is resampled in a slow time dimension, as expressed by: in, This is the broadband observation data after resampling correction. This is the original array multi-channel broadband observation data. For observation data in the fast time-frequency domain; For slow time, Indicates about fast time Fourier transform, Indicates frequency with fast time Inverse Fourier transform, The slow time coordinates are after resampling; Step 2.2: In the resampled and corrected broadband observation data Above, a distance dimension statistic is formed. The expression is: in, For slow-time sets, The magnitude of the complex amplitude; Introducing a distance-dimensional adaptive threshold right Local peak extraction is performed to obtain a set of candidate distance gates. And the corresponding distance, expressed as: in, For the first One candidate distance gate; This is a local peak operator used to extract representative distance peaks from a continuous region that passes through a threshold; For the first Distance estimation for each candidate distance gate; The speed of light; If the candidate distance gate set is not empty, calculate the target distance corresponding to the peak value; otherwise, determine that there is no target in this frame and wait for the next frame's observation input. Step 2.3: For candidate distance gates ,exist The array observation vector is extracted and a angular domain response is formed on a preset scanning angular grid to determine the coarse-angle unit corresponding to the main response peak, which serves as the initial estimate of the target angle. Within the main lobe neighborhood corresponding to the coarse-angle unit, multiple adjacent scanning beam directions are selected, and the scanning beams at the candidate range gate are extracted. The complex observation output at the specified location is coherently accumulated in the slow time dimension to construct a multibeam coherent observation vector: in, Indicates the first Frames in candidate distance gate Place, No. Coherent observations along the scanning beam direction; , The number of adjacent scan beams selected; Indicates the first Frames in candidate distance gate Multibeam coherent observation vector at location; superscript It is the transpose symbol; Based on the geometric parameters of the radar array, the target's position at a given angle is established. The theoretical response vector of the multibeams under the following conditions for: in, These are the target azimuth and elevation angles, respectively. For the first The theoretical response value in each scanning beam direction, ; No. The theoretical response in each scanning beam direction can be expressed as a coherent superposition of the array direction responses as follows: in, , These represent the number of array elements in the horizontal and vertical directions, respectively. These are the array element numbers for the horizontal and vertical directions, respectively. For the array element space coordinates, For the first Each scanning beam array element The corresponding weighting coefficients; Define the multibeam coherence matching metric function as follows: in, The value of the multibeam coherent matching metric function. for The conjugate transpose of; The matching metric function eliminates the influence of differences in gain between different scanning beams and echo amplitude fluctuations on the angle measurement results through energy normalization; a local search is performed within the coarse angle cell corresponding to the main response peak to determine the angle parameter that maximizes the matching metric function. This refers to the estimated results of the azimuth and elevation angles; Step 2.4, for the first Candidate distance gate for frames The obtained distance and angle estimates are combined to obtain a joint measurement vector: in, For the first Frame number The joint measurement vector of the candidate distance gates, , , The first Frame number Distance estimation, azimuth estimation, and pitch estimation for each candidate distance gate; For each joint measurement A reasonableness assessment is conducted, based on pre-defined constraints and consistency requirements, to determine whether the joint measurement meets physical feasibility, observation consistency, and output reliability; when the joint measurement... If the reasonableness condition is not met, the joint measurement is deemed unreliable and removed; if the reasonableness condition is met, the joint measurement is retained and added to the output set. .
3. The method for extended target detection and tracking of UAV swarms based on broadband radar according to claim 2, characterized in that, Step 3, which involves performing data association based on joint information, introduces the spatial shape parameters of the cluster targets as state variables into the tracking model. These parameters are then used in conjunction with the target motion state for prediction and updating, enabling dynamic control of the candidate observation range and improving the continuity and robustness of subsequent tracking. Specifically, the steps are as follows: Step 3.1, place the first The set of joint measurements retained after the frame passes the rationality judgment is denoted as . ,right Intra-frame point merging is performed, defining intra-frame nearest neighbor criteria based on three dimensions: distance, azimuth, and pitch. Measurements satisfying the nearest neighbor relationship are grouped into the same cluster, and the intra-frame candidate point set of this cluster is: , Indicates the first Frame number The number of merged points within a cluster is used to generate the corresponding number of points in the cluster statistical representative value. Frame number Fusion measurement of individual clusters ; Will Translated to Cartesian coordinates, the three-dimensional Cartesian uniform state and cluster shape measurement matrix are defined as follows: in, For state estimation, The state matrix; For the first Frame number A shape measurement matrix for each cluster, used to characterize the spatial scale and shape of the cluster; subscript Indicates frame number and index Indicates the cluster sequence number; Indicates the first Frame number Within the cluster Cartesian coordinate vectors of points; It is a positive number less than a set value, used to prevent the matrix from being non-positive definite; It is a three-dimensional identity matrix; To achieve stable cross-frame correlation, the first frame... Shape-aware gating is set for the predictive prior of each cluster: in, For the first Frame number The predicted state value of each cluster. For the first Frame number The state prediction covariance matrix of each cluster The noise covariance is measured as a basis. For the first The shape prediction matrix of each cluster The number of valid points is estimated based on data from neighboring frames. This is the gate threshold; Step 3.2: After completing the cross-frame point association, adaptive measurement noise for point count and shape distribution is introduced. The expression for the state update process is: in For the first Frame number Adaptive measurement noise covariance of each cluster; For the innovation covariance in the update phase, For the first The state filter estimate after frame update For the first Covariance of the state filter estimate after frame update; The filter gain is used for an adaptive trade-off between the predicted prior and the current observation; This is the shape smoothing coefficient, used to weight the shape prediction and shape measurement. The identity matrix is consistent with the dimension of the state; Step 3.3: To perform gating association and maintain trajectory continuity in the next frame, based on the inter-frame time interval... The state is predicted, and the shape matrix is predicted by random walk and positive definite constraint processing is applied. The expressions for the state transition matrix and the prediction process are as follows: in, This is the state transition matrix; For the next frame, predict the filtered estimate. The covariance estimated by the prediction filter for the next frame. For the first The shape prediction matrix for the next frame of each cluster. For state process noise covariance, For shape process noise covariance, This involves symmetricizing the matrix and imposing a lower bound constraint on the eigenvalues to ensure that the result is a symmetric positive definite matrix. Indicates the next frame prediction fusion measurement; Step 3.4: To reduce the search range in the next frame and make the search window adapt to the cluster scale, a predicted measurement scattering is constructed based on shape prediction, and a spherical coordinate domain window is generated. in, For the first Frame number The search window vector for each cluster, The width of the search window is the distance dimension. The width of the azimuth search window. The width of the search window is the pitch angle dimension.
4. The method for extended target detection and tracking of UAV swarms based on broadband radar according to claim 3, characterized in that, In step 3.1, for each fused measurement in the current frame, if the fused measurement passes the prediction gate of any existing trajectory, then the fused measurement is associated with the corresponding trajectory; if the fused measurement does not pass the prediction gate of any existing trajectory, then the fused measurement is used as the initial measurement for the new trajectory; for trajectories that are not associated with any measurement in multiple consecutive frames, they are maintained or terminated according to a preset strategy.
5. The method for extended target detection and tracking of UAV swarms based on broadband radar according to claim 4, characterized in that, In step 3, to achieve cyclical data updates, the prediction fusion measurement is... The gating center is used for step 3.1 in the next frame. At the same time, the shape prediction matrix and prediction covariance adaptively determine the scale and direction of the gating, so as to automatically widen the gating range when the cluster scale expands and automatically tighten the gating range when the cluster scale shrinks.
6. The method for extended target detection and tracking of UAV swarms based on broadband radar according to claim 5, characterized in that, Step 3 further forms a continuous closed-loop processing chain: At the beginning of each frame processing, the estimation results of the previous frame are used to generate the prior prediction of the next frame, adaptively limiting the candidate observation range and completing the gating screening; cross-frame association is performed on the candidate observations that pass the gating to form the effective overall observation input of the current processing frame, and the overall state and shape of the cluster are recursively updated; the updated state is then used to generate the prediction prior of the next frame and fed back to the subsequent gating and association links, realizing frame-by-frame output and improving the continuity and robustness of subsequent tracking.
7. A wideband radar-based extended target detection and tracking system for unmanned aerial vehicle (UAV) swarms, characterized in that, This system is used to implement the broadband radar-based extended target detection and tracking method for UAV swarms as described in any one of claims 1 to 6, the system comprising: The broadband radar modeling module for cluster targets sets up a UAV cluster scenario, calculates the electromagnetic scattering characteristics of the target based on the scenario, and equates the UAV cluster to an extended target composed of multiple scattering centers. It then combines noise and interference modeling to generate broadband radar echoes of the cluster extended target. The cluster target broadband radar detection module calibrates and compensates the broadband observation data. If no candidate target is found, it waits for the next frame of data input. When the target distance information is extracted in the distance dimension, the multi-beam coherent fitting method is used to estimate the angle, and the cluster target distance-angle detection result is jointly output. The cluster target broadband radar tracking module performs data association based on joint information, introduces the spatial shape parameters of the cluster targets as state variables into the tracking model, and predicts and updates them together with the target motion state, thereby realizing dynamic control of the candidate observation range and improving the continuity and robustness of subsequent tracking.
8. A mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the wideband radar-based extended target detection and tracking method for UAV swarms as described in any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the broadband radar-based UAV swarm extended target detection and tracking method as described in any one of claims 1 to 6.