Bird repelling method using low-altitude radar combined with high-point monitoring

Abstract

The present invention relates to the field of aviation safety technology, specifically a method for bird repelling using low-altitude radar combined with high-point monitoring. The method comprises the following steps: using a low-altitude radar deployed at a matrix node at the airport boundary, acquiring raw data of airborne reflection echoes in reconnaissance mode, constructing a three-dimensional perception spatial point set including detection time, target slant range, azimuth angle, pitch angle, and echo intensity; and employing an improved DeepSORT technique to intelligently identify the spatial point set in step S1. The present invention constructs a three-dimensional perception network using low-altitude radar at the airport boundary, and employs an improved DeepSORT technique for secondary verification, achieving a bird recognition accuracy rate of ≥98% and a positioning refresh rate of <1 second. To address the limited video surveillance coverage, the radar constructs a three-dimensional perception network to guide camera focus, an EMA module enhances small target feature extraction, a linked drone dynamically plans a path, and carries multimodal equipment for repelling birds. A reinforcement learning strategy is employed, significantly reducing the rate of repeated bird intrusions.

Description

Technical Field

[0001] The present invention relates to the field of aviation safety technology, in particular to a bird-repelling method using a low-altitude radar combined with high-point monitoring. Background Art

[0002] Bird strike prevention at airports is an important component of the aviation safety system. Bird strikes can cause serious consequences such as engine damage and windshield rupture, threatening not only the lives of passengers and crew members but also significant economic losses. Therefore, it is particularly important to develop efficient airport bird-repelling technology and establish a real-time bird flock prevention and control system. Modern bird-repelling technology has evolved from traditional passive defense to an intelligent, integrated prevention and control system that includes low-altitude radar detection, optical recognition, and acoustic interference. Its core technology lies in breaking the adaptive laws of bird behavior and achieving sustainable ecological safety protection. Therefore, it is necessary to establish a modern, three-dimensional intelligent protection network for airport airspace protection zones, using various technical means to prevent birds from entering the airport flight area and effectively reduce the risk of collision between aircraft and birds.

[0003] For example, the patent with announcement number CN118411027A discloses an active bird-strike prevention method and device for airports. The bird-strike prevention method includes the following steps: collecting bird information; comparing and analyzing the collected bird information with the bird work management database; and sending bird-repelling information to the corresponding bird-repelling equipment based on the analysis results to repel birds. The present invention provides a comprehensive bird-repelling system that integrates low-altitude bird-detecting radar and bird-repelling linkage by setting up a detection and drive system. The bird-repelling equipment is deployed in a combined decentralized manner to cover the airport perimeter in all directions. Relying on the smart bird prevention platform, the deployment position of the bird-repelling equipment can be flexibly adjusted according to the activity patterns of birds near the airport to prevent birds from having an adaptive impact on the equipment. Therefore, based on the existing core area safety goals and performance indicators, with the help of various currently available bird-repelling means, the detection and drive system is used to ensure that birds are immediately dispersed once they approach the core area.

[0004] However, when the above technologies are actually used, there are problems such as high false alarm rate of low-altitude radar, limited video surveillance coverage, and lack of coordinated linkage between monitoring and expulsion equipment. Summary of the Invention

[0005] The purpose of the present invention is to provide a low-altitude radar combined with high-point monitoring bird repelling method to solve the problems of high false alarm rate of low-altitude radar, limited video monitoring coverage, and lack of coordinated linkage between monitoring and repelling equipment.

[0006] To achieve the above object, the present invention provides the following technical solution: a low-altitude radar combined with high-point monitoring bird-repelling method, comprising the following steps:

[0007] S1. Low-altitude radar scanning initial data set: Low-altitude radars deployed at the airport boundary matrix nodes acquire raw data of air reflection echoes in reconnaissance mode, and construct a three-dimensional perception space point set containing detection time, target slant range, azimuth angle, pitch angle, and echo intensity.

[0008] S2. Initial bird flock identification: Using the improved DeepSORT technology to intelligently identify the spatial point set in step S1, a tracking trajectory for an aerial mobile target is formed. The improvements include: adding the echo intensity dimension s and the height parameter w to the Kalman filter state variable, and using Mahalanobis distance and cosine distance weighted metrics for trajectory association to achieve tracking and prediction of the target trajectory;

[0009] S3. Bird flock identification and verification: The tracking trajectory identified in step S2 is combined with the low-altitude radar coordinates to calculate the WGS-84 coordinates. The infrared dual-band smart camera is used for target verification. The camera verification algorithm is based on Yolov8s. An efficient multi-scale attention module (EMA) is added after the second convolutional layer of the Backbone C2f module. Together with C2f, it forms a C2f-EMA module, replacing the original Backbone third- and fourth-level C2f modules to improve the feature extraction capabilities of small and blurred targets.

[0010] S4, Tracking mode activated: After the low-altitude radar and camera confirm the track in S3, the low-altitude radar switches to tracking mode, and the bird track update interval is shortened from 3 seconds to 0.1 seconds;

[0011] S5. Collaborative decision-making and task dispatch: Based on the bird flock location, flight direction, and airport protection zone data from step S4, combined with the heading angle change rate and velocity vector of the bird flock's historical trajectory, a short-term prediction of the bird flock's movement trajectory within the next 30 seconds is made. Based on the predicted trajectory from step S2, the minimum distance between the bird flock and the protection zone is predicted. If the minimum distance is less than the threshold, a bird-scaring mission is generated and the drone is dispatched.

[0012] S6. Dynamic path planning and bird-repelling execution: When the bird-repelling mission is in progress, the low-altitude radar provides real-time guidance for the UAV. The UAV plans the path based on the real-time location of the bird flock and carries out the repelling mission with multi-modal bird-repelling equipment. The minimum distance between the bird flock and the protected area is calculated simultaneously. When the distance meets the threshold and the predicted trajectory no longer enters the protected area, the low-altitude radar returns to the reconnaissance mode.

[0013] Preferably, in step S2, the echo intensity dimension s and the height parameter w are added to the Kalman filter state quantity, and the calculation formula is:

[0014] X k =AX k-1 +BU k-1 +w k-1 ;

[0015] Where A: state transfer matrix;

[0016] B: control matrix;

[0017] U: control input;

[0018] X: state, whose parameters are {u,v,w,r,h,s,u′,v′,w′,r′,h′,s′};

[0019] Where u, v, w: target center coordinates;

[0020] r: aspect ratio;

[0021] h: target height;

[0022] s: echo intensity dimension;

[0023] u′,v′,w′,r′,h′,s′: the speed corresponding to each parameter.

[0024] Preferably, the formula for weighted measurement of Mahalanobis distance and cosine distance in step S2 is:

[0025] D=λd 马氏 +(1-λ)d 余弦 ;

[0026] Where D is the comprehensive measurement result;

[0027] λ is the weighting factor;

[0028] d 马氏 is the Mahalanobis distance measurement result;

[0029] d 余弦 is the cosine distance measurement result.

[0030] Preferably, in step S2, the DeepSORT technology divides the trajectories into deterministic trajectories and non-deterministic trajectories, the deterministic trajectories are associated using a combination of Mahalanobis distance and cosine distance metrics, and the non-deterministic trajectories are associated using an IOU method.

[0031] Preferably, the position update formula of the bird flock in step S4 can be described as:

[0032]

[0033] in is the flock velocity vector, obtained by DeepSORT;

[0034] dτ is a small increment of time;

[0035] is the current position vector;

[0036] is the position vector at the initial moment.

[0037] Preferably, the drone path planning formula in step S6 is:

[0038]

[0039] where υ u is the speed of the drone;

[0040] is the distance ‖d(t)‖ between the drone and the flock of birds;

[0041] is the drone position vector;

[0042] Δt is the time difference between adjacent moments.

[0043] Preferably, the multimodal bird-repelling device in step S6 includes switching of sound wave, laser, and airflow disturbance modes.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This invention applies DeepSORT to low-altitude radar for the first time, and adds the echo intensity dimension s and height parameter w to the Kalman filter in DeepSORT. Compared with traditional radars that use algorithms such as Kalman filtering, DeepSORT has a better target tracking effect, achieving a bird recognition accuracy rate of ≥98% and a positioning refresh rate of <1 second. In view of the limited video surveillance coverage, the radar builds a three-dimensional perception network to guide the camera focus, the EMA module enhances the feature extraction of small targets, and the drone is linked to dynamically plan the path and equipped with multi-modal equipment to drive away birds. The reinforcement learning strategy greatly reduces the repeated bird intrusion rate. At the same time, the improved Yolov8s algorithm is better than the standard Yolov8s algorithm in recognizing small targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of an improved low-altitude radar target tracking algorithm for the low-altitude radar combined with high-point monitoring and bird-repelling method of the present invention;

[0047] Figure 2 Schematic diagram of the improved Yolov8 algorithm model for the low-altitude radar combined with high-point monitoring bird-repelling method of the present invention;

[0048] Figure 3 This is a schematic diagram of the EMA attention algorithm structure for improving the small target recognition rate of the low-altitude radar combined with high-point monitoring bird-repelling method of the present invention;

[0049] Figure 4 This is a schematic diagram of a flock of birds captured by a camera in an embodiment of the low-altitude radar combined with high-point monitoring bird-repelling method of the present invention;

[0050] Figure 5 This is a schematic diagram of the movement trajectory of a flock of birds tracked by a low-altitude radar in an embodiment of the low-altitude radar combined with high-point monitoring bird-repelling method of the present invention;

[0051] Figure 6 This is a schematic diagram of the movement trajectory of a flock of birds and a drone in an embodiment of the low-altitude radar combined with high-point monitoring bird-repelling method of the present invention;

[0052] Figure 7 1 is a schematic diagram comparing the effects of the improved DeepSORT tracking algorithm used in the present invention and the conventional Kalman filtering method;

[0053] Figure 8 2 is a schematic diagram comparing the effects of the improved DeepSORT tracking algorithm used in the present invention and the conventional Kalman filtering method;

[0054] Figure 9 3 is a schematic diagram comparing the effects of the improved DeepSORT tracking algorithm used in the present invention and the conventional Kalman filtering method;

[0055] Figure 10 This is a schematic diagram of the improved Yolov8s used in the present invention for identifying small target birds;

[0056] Figure 11 This is a schematic diagram of the standard Yolov8s for small target bird recognition. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] See also Figure 1-11 The present invention provides a technical solution: a low-altitude radar combined with high-point monitoring bird-repelling method, comprising the following steps:

[0059] S1. Low-altitude radar scanning and bird flock identification: Low-altitude radar deployed near the airport uses reconnaissance mode to obtain the original trace information of various air reflection echoes. The trace information includes detection time, target slant range, azimuth angle, pitch angle, echo intensity, etc.

[0060] S2. Use DeepSORT technology to intelligently identify the spatial point set scanned by the low-altitude radar to form a tracking trajectory for the movable target in the air. Figure 1As shown in the figure, during the low-altitude radar scanning process, the low-altitude radar original point track data is sent to DeepSORT as input information. The recognition algorithm divides the trajectories into deterministic trajectories and non-deterministic trajectories. For deterministic trajectories, the Mahalanobis distance and cosine distance metrics are combined to achieve trajectory association. For non-deterministic trajectories, the IOU method is used to determine the association.

[0061] Improvement of Kalman filter: For trajectory tracking and prediction, the algorithm adds the echo intensity dimension as one of the state quantities based on the original Kalman filter state quantity. At the same time, since the original DeepSORT is used for two-dimensional image tracking, in order to adapt to three-dimensional target scenes, the center coordinate adds the height parameter w as one of the state quantities, that is:

[0062] X k =AX k-1 +BU k-1 +w k-1 ;

[0063] Where A: state transfer matrix;

[0064] B: control matrix;

[0065] U: control input;

[0066] X: state, whose parameters are {u,v,w,r,h,s,u′,v′,w′,r′,h′,s′}, u,v,w: target center coordinates; r: aspect ratio; h: target height; s: echo intensity dimension; u′,v′,w′,r′,h′,s′: the speed corresponding to each parameter.

[0067] (1) Mahalanobis distance measurement method. This method is used to measure the distance between the trajectory predicted by the Kalman filter and the new input trajectory, which takes into account the uncertainty of state estimation. This method is mainly used to determine the correlation between consecutive frames. Its calculation formula is as follows:

[0068]

[0069] S i : target covariance matrix predicted by i Kalman filters;

[0070] d j : The jth target position detected;

[0071] y i : The target position predicted by the i-th Kalman filter;

[0072] The smaller d(i,j) is, the greater the correlation between the trajectory and the original data is.

[0073] (2) Cosine distance measurement method. This method is based on appearance information. During low-altitude radar detection, the echo of a bird usually cannot properly reflect the shape characteristics of the bird itself. At the same time, due to electromagnetic interference, the shapes of two adjacent echoes of the same target may be quite different. Therefore, the cosine distance is used.

[0074] This algorithm combines the strengths of Mahalanobis distance and cosine distance to provide solutions for both long-term and short-term target tracking. The Mahalanobis distance measures the difference in the position of a target object between two consecutive frames, addressing the problem of tracking targets between consecutive frames. The cosine distance compares the similarity between the features of a newly detected object and those of an already tracked object, addressing the problem of tracking targets between intervening frames. To address the association problem, DeepSORT combines these two metrics through a weighted summation, as shown below.

[0075] D=λd 马氏 +(1-λ)d 余弦 ;

[0076] Where D is the comprehensive measurement result, d 马氏 is the Mahalanobis distance measurement result, d 余弦 is the cosine distance measurement result, and λ is the weighting factor.

[0077] S3. Based on the trajectory of the movable target identified by the low-altitude radar, the WGS-84 coordinate position of the target is calculated in combination with the coordinate position of the low-altitude radar itself. The target position is sent to the infrared dual-band intelligent camera for further target identification. The camera target recognition algorithm is based on Yolov8s. Considering the small proportion of bird target pixels, the EMA efficient multi-scale attention module is added to Backbone. Figure 3 shown.

[0078] This module is located after the second convolutional layer of the C2f module and is used to receive data after convolution and dimension increase. Together with C2f, it forms the C2f-EMA module, replacing the third and fourth level C2f modules in the original Backbone. Figure 2 As shown in the figure, it can effectively improve the model's ability to extract small and fuzzy target features.

[0079] S4. Tracking mode activated: After the low-altitude radar and camera simultaneously confirm the bird's trajectory, the low-altitude radar switches from reconnaissance mode to tracking mode, continuously detecting birds. The update interval of the bird's trajectory is increased from 3 seconds to 0.1 seconds.

[0080] S5. Collaborative decision-making and task dispatch: Based on the bird position and flight direction in S4, combined with the airport protection zone data, and the heading angle change rate and velocity vector of the bird flock's historical trajectory, a short-term prediction of the bird flock's movement trajectory in the next 30 seconds is made. Based on the predicted trajectory in step S2, the minimum distance between the bird flock and the protection zone is predicted (the threshold setting rule is: the basic threshold is 500 meters, and it can be adjusted according to user needs: for large bird flocks or high-speed flying bird flocks, the threshold is increased to 800 meters). If it is less than the threshold, a bird-scaring task is generated and the drone is dispatched.

[0081] The latitude and longitude data of the bird flock fed back by the low-altitude radar are expressed as the position of the dynamic target, that is, a point in the three-dimensional coordinate system. The position update formula of the bird flock can be described as:

[0082]

[0083] The current latitude and longitude position of the flock:

[0084] dτ is a small increment of time;

[0085] Initial position of the flock:

[0086] Velocity vector of the flock: The direction is the flight direction of the flock of birds, and the position and speed can be obtained by DeepSORT.

[0087] If the flock position is updated discretely (via periodic feedback from a low-altitude radar), it can be approximated by a difference equation:

[0088]

[0089] The drone adjusts its flight path based on the latest location of the flock, aiming to approach the flock and drive it away. The drone's path planning can be based on the following formula:

[0090] The current location of the drone is The target position is set to the current position of the flock Then the target direction vector of the UAV is:

[0091]

[0092] Drone speed: υ u

[0093] Distance between drone and flock of birds:

[0094]

[0095] To avoid direct collisions between the drone and the flock of birds while ensuring close proximity, direction adjustment can be added with obstacle avoidance or approach restrictions, such as setting a minimum safe distance:

[0096]

[0097] S6. Dynamic Path Planning and Bird Repelling Execution: After the bird repelling command is issued in step S5, the low-altitude radar provides real-time guidance to the drone using tracking mode. The drone plans a path based on the real-time location of the flock and carries out the repelling mission using multimodal bird repelling equipment. The minimum distance between the birds and the flight is simultaneously calculated. The bird repelling mission ends when the minimum distance between the birds and the airport protection zone meets the threshold and the low-altitude radar predicts that the birds will no longer enter the airport protection zone. The low-altitude radar then switches to reconnaissance mode.

[0098] This system combines the sensing capabilities of a low-altitude phased array radar with a dual-band infrared intelligent camera, optimizing multiple algorithms to build a three-dimensional perception network. This improves the accuracy and timeliness of bird flock identification. The system employs a multimodal data fusion algorithm based on deep learning to simultaneously analyze key parameters such as bird species, numbers, and flight paths. Testing has shown a bird identification accuracy rate of ≥98%, with a positioning refresh rate of less than 1 second.

[0099] The present invention establishes a closed-loop management of intelligent decision-making and adaptive optimization. Through the risk assessment model, it combines real-time bird flock data and flight dynamics to accurately quantify the bird strike risk and realize the automatic dispatch of bird-scaring tasks.

[0100] The present invention deploys low-altitude radars at the airport boundary matrix nodes, obtains the original data of air reflection echoes in the reconnaissance mode to construct a three-dimensional perception space point set, adopts the improved DeepSORT technology (adds the echo intensity dimension and height parameter to the Kalman filter state quantity, and adopts Mahalanobis distance and cosine distance weighted measurement for trajectory association) to form the tracking trajectory of the movable target in the air, and pulls the infrared dual-band intelligent camera based on Yolov8s and adding EMA module in Backbone to verify the target trajectory. After the radar and the camera confirm, they switch to the tracking mode to shorten the trajectory update interval from 3 seconds to 0.1 second. The bird-repelling task is generated by combining the bird flock position, flight direction and airport protection zone data. The low-altitude radar can effectively detect the target trajectory and effectively avoid the bird-repelling task. The guided drone plans the path based on the real-time position of the bird flock and carries acoustic, optical, and airflow disturbance multi-modal bird-repelling equipment to perform the repelling task until the threshold requirements are met; its technical feature is that it combines the perception capabilities of low-altitude phased array radar and infrared dual-band intelligent camera, and realizes the quantification of bird strike risk and automatic dispatch of bird-repelling tasks through multi-modal data fusion algorithm (recognition accuracy ≥ 98%, positioning refresh frequency < 1 second) and risk assessment model, and uses reinforcement learning model to dynamically adjust the bird-repelling strategy. The measured bird repeat invasion rate is reduced by 83%, and a high-precision and highly coordinated intelligent bird-repelling system is constructed, which solves the problems of high radar false alarm rate, insufficient recognition of small targets by video surveillance, and lack of coordinated linkage of equipment in traditional bird-repelling systems.

[0101] The present invention was tested and verified in a resident's yard in Nanhuaidian Village, Ninghe District, Tianjin on June 3, 2025. Compared with the traditional radar using algorithms such as Kalman filtering, DeepSORT has a better tracking effect on the target. Figure 7 、 Figure 8 、 Figure 9 As shown in the figure, the red markers represent the original track points reported by the low-altitude radar, the blue markers represent the bird flight paths fused using the improved DeepSORT method adopted by this invention, and the orange markers represent the bird flight paths fused directly using the Kalman filter algorithm. As can be seen from the figure, the improved DeepSORT algorithm outperforms the traditional radar Kalman filter algorithm in terms of track smoothness.

[0102] The improved Yolov8s used in this paper has the following recognition effects on small targets: Figure 10 As shown in the figure, the standard Yolo8s’s bird recognition performance in the same environment is as follows: Figure 11 As shown in the figure, it can be seen that the improved Yolov8s recognition algorithm has better detection effect on small targets than the standard Yolo8s algorithm.

[0103] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0104] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The method of bird-repelling by low-altitude radar combined with high-point monitoring is characterized in that: The following steps are involved: S1. Low-altitude radar scanning initial data set: Low-altitude radars deployed at the airport boundary matrix nodes acquire raw data of air reflection echoes in reconnaissance mode, and construct a three-dimensional perception space point set containing detection time, target slant range, azimuth angle, pitch angle, and echo intensity. S2. Initial bird flock identification: Using the improved DeepSORT technology to intelligently identify the spatial point set in step S1, a tracking trajectory for an aerial mobile target is formed. The improvements include: adding the echo intensity dimension s and the height parameter w to the Kalman filter state variable, and using Mahalanobis distance and cosine distance weighted metrics for trajectory association to achieve tracking and prediction of the target trajectory; S3. Bird flock identification and verification: The tracking trajectory identified in step S2 is combined with the low-altitude radar coordinates to calculate the WGS-84 coordinates. The infrared dual-band smart camera is used for target verification. The camera verification algorithm is based on Yolov8s. An efficient multi-scale attention module (EMA) is added after the second convolutional layer of the Backbone C2f module. Together with C2f, it forms a C2f-EMA module, replacing the original Backbone third- and fourth-level C2f modules to improve the feature extraction capabilities of small and blurred targets. S4, Tracking mode activated: After the low-altitude radar and camera confirm the track in S3, the low-altitude radar switches to tracking mode, and the bird track update interval is shortened from 3 seconds to 0.1 seconds; S5. Collaborative decision-making and task dispatch: Based on the bird flock location, flight direction, and airport protection zone data from step S4, combined with the heading angle change rate and velocity vector of the bird flock's historical trajectory, a short-term prediction of the bird flock's movement trajectory within the next 30 seconds is made. Based on the predicted trajectory from step S2, the minimum distance between the bird flock and the protection zone is predicted. If the minimum distance is less than the threshold, a bird-scaring mission is generated and the drone is dispatched. S6. Dynamic path planning and bird-repelling execution: When the bird-repelling mission is in progress, the low-altitude radar provides real-time guidance for the UAV. The UAV plans the path based on the real-time location of the bird flock and carries out the repelling mission with multi-modal bird-repelling equipment. The minimum distance between the bird flock and the protected area is calculated simultaneously. When the distance meets the threshold and the predicted trajectory no longer enters the protected area, the low-altitude radar returns to the reconnaissance mode.

2. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: In step S2, the echo intensity dimension s and the height parameter w are added to the Kalman filter state quantity, and the calculation formula is: X k =AX k-1 +BU k-1 +w k-1 ; Where A: state transfer matrix; B: control matrix; U: control input; X: state, whose parameters are {u,v,w,r,h,s,u′,v′,w′,r′,h′,s′}; Where u, v, w: target center coordinates; r: aspect ratio; h: target height; s: echo intensity dimension; u′,v′,w′,r′,h′,s′: the speed corresponding to each parameter.

3. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: The formula for weighted measurement of Mahalanobis distance and cosine distance in step S2 is: D=λd 马氏 +(1-λ)d 余弦 ; Where D is the comprehensive measurement result; λ is the weighting factor; d 马氏 is the Mahalanobis distance measurement result; d 余弦 is the cosine distance measurement result.

4. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: In step S2, the DeepSORT technology divides the trajectories into deterministic trajectories and non-deterministic trajectories. The deterministic trajectories are associated using a combination of Mahalanobis distance and cosine distance metrics, while the non-deterministic trajectories are associated using the IOU method.

5. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: The formula for updating the position of the flock of birds in step S4 can be described as: in is the flock velocity vector, obtained by DeepSORT; dτ is a small increment of time; is the current position vector; is the position vector at the initial moment.

6. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: The UAV path planning formula in step S6 is: where v u is the speed of the drone; is the distance ‖d(t)‖ between the drone and the flock of birds; is the drone position vector; Δt is the time difference between adjacent moments.

7. The low-altitude radar combined with high-point monitoring bird-repelling method according to claim 1 is characterized in that: The multi-modal bird-repelling device in step S6 includes switching between sound wave, laser, and airflow disturbance modes.

Images