Collaborative awareness and identification method and system based on unmanned aerial vehicle cluster

Abstract

The invention discloses a collaborative awareness and recognition method and system based on an unmanned aerial vehicle cluster, and relates to the field of unmanned aerial vehicle cluster control and computer vision. The method comprises the steps of starting ground control software to load a route path file, constructing a sequence image optical scene and evaluating the effect, dynamically fusing and evaluating the effect of a wide-area optical scene, intelligently and collaboratively recognizing an optical target, confirming and judging collaborative recognition data, and counting and evaluating recognition accuracy. The system comprises a central node unmanned aerial vehicle subsystem, an edge node unmanned aerial vehicle subsystem and ground control software. According to the method, the central node and the edge nodes cooperatively work, the global scheduling decision-making capability of the central node and the real-time computing capability of the edge nodes are exerted, efficient coverage of a wide-area scene, high-resolution image splicing and cooperative recognition and tracking of a dynamic target are achieved, the problems of unbalanced computing load, information redundancy and the like of a traditional architecture are solved, and the method is suitable for large-scale popularization and application. The image quality and the target recognition accuracy are improved, the reliability is high, and the expandability is high.

Description

Technical Field

[0001] The present invention relates to the fields of drone cluster control and computer vision, and in particular to a collaborative perception and recognition method and system based on drone clusters. Background Art

[0002] With the rapid development of drone technology, drone swarm systems are increasingly being used in military reconnaissance, disaster monitoring, environmental assessment, and other fields. Traditional drone systems typically employ centralized or distributed architectures for wide-area scene perception and target recognition, but both models have significant limitations. Centralized architectures rely on a single central node for data processing and decision-making. While they can achieve global information fusion, the excessive computational load on the central node can easily become a performance bottleneck. If the central node fails, the entire system faces the risk of paralysis. While distributed architectures share computing tasks across multiple nodes, they lack an efficient global scheduling mechanism, which can easily lead to information redundancy or omissions, making efficient collaborative perception and recognition difficult.

[0003] In the military, real-time monitoring and target identification of wide-area scenarios are crucial for battlefield situational awareness. For example, in battlefield reconnaissance, drone swarms must rapidly identify and track dynamic targets (such as vehicles, personnel, and equipment) across a large area. However, traditional centralized architectures, due to the limited computing power of central nodes, struggle to process the massive amounts of sensory data in wide-area scenarios in real time, resulting in inadequate timeliness and accuracy in target identification. While distributed architectures can share computing tasks, they lack a global information fusion mechanism, making efficient collaboration among multiple nodes difficult, which can easily lead to missed or false detections of targets.

[0004] In the civilian sector, drone swarm systems also demonstrate significant potential in scenarios such as disaster monitoring and environmental assessment. For example, after natural disasters like earthquakes and floods, drone swarms can rapidly conduct large-scale scans of the affected area, acquiring high-resolution image data to support disaster assessment and rescue decision-making. However, traditional centralized architectures, due to limited computing resources at central nodes, struggle to process complex data across wide areas in real time, resulting in ineffective disaster assessments. While distributed architectures can share computing tasks, the lack of an efficient global scheduling mechanism can easily lead to task conflicts and information redundancy among multiple drones, compromising detection efficiency.

[0005] In recent years, researchers have proposed a variety of drone collaborative perception and recognition methods, but most have failed to effectively address the coordination issues between central and edge nodes. For example, while distributed computing-based drone swarm systems improve computational efficiency through task allocation, they lack global scheduling and information fusion mechanisms at the central node, making efficient collaborative perception and recognition difficult to achieve. Furthermore, while deep learning-based drone target recognition methods have improved recognition accuracy to a certain extent, the high computational complexity of the models makes them difficult to run in real time at edge nodes. Therefore, designing a drone swarm system based on a collaborative model between central and edge nodes to achieve efficient perception and target recognition in wide-area scenarios is a key research direction in the current technology landscape. Summary of the Invention

[0006] To solve the above technical problems, the present invention proposes a collaborative perception and recognition method and system based on drone clusters. Through the collaborative operation of the central node and edge nodes, the system can give full play to the global scheduling and decision-making capabilities of the central node, while utilizing the real-time computing capabilities of the edge nodes to achieve efficient coverage of wide-area scenes, high-resolution image stitching, and collaborative recognition and tracking of dynamic targets, providing strong support for real-time monitoring and decision-making in complex environments.

[0007] A collaborative perception and recognition method based on drone clusters specifically includes the following steps:

[0008] Step S1: Start the ground control software and load the UAV route path file;

[0009] Step S2: The central node UAV subsystem receives the image data captured by the edge node UAV subsystem, constructs a sequence image optical scene using the algorithm dynamic library carried by the central node UAV subsystem, and evaluates the effect of the wide-area optical scene construction;

[0010] Step S3: The edge node UAV subsystem collects high-resolution image data and sends it to the central node UAV subsystem. The central node UAV subsystem uses the algorithm dynamic library to perform wide-area optical scene dynamic fusion and evaluate the wide-area optical scene dynamic fusion effect.

[0011] Step S4: The central node UAV subsystem preliminarily identifies the target drone and guides the edge node UAV subsystem. The edge node UAV subsystem feeds back the high-resolution recognition results to the central node UAV subsystem, which then performs decision-level fusion recognition to achieve intelligent collaborative recognition of optical targets.

[0012] Step S5: Collaborative identification data confirmation and determination;

[0013] Step S6: Recognition accuracy statistics and evaluation.

[0014] Furthermore, in step S2, the construction and effect evaluation of the optical scene of the sequence of images specifically include the following steps:

[0015] Step S2-1: The central node UAV and the edge node UAV fly according to the cluster control instructions of the ground control software and perform field detection to collect image data;

[0016] Step S2-2: The central node UAV processor prototype receives image data captured by the edge node UAV subsystem, constructs a sequence image optical scene using the sequence image optical scene construction algorithm dynamic library installed on the central node processor prototype, and transmits the constructed optical scene image to the ground control software for display;

[0017] Step S2-3: Evaluate the wide-area optical scene image effect constructed by the central node processor prototype and the dynamic library of the sequential image optical scene construction algorithm it carries. The evaluation indicators include visual distortion and splicing misalignment.

[0018] Furthermore, in step S3, the dynamic fusion and effect evaluation of the wide-area optical scene specifically includes the following steps:

[0019] Step S3-1: The edge node drone improves the field of view image resolution by zooming and collects high-resolution image data;

[0020] Step S3-2: The edge node UAV subsystem sends the high-resolution image data to the central node UAV subsystem. The high-resolution image is dynamically fused to the corresponding area of ​​the wide-area optical scene using the wide-area optical scene dynamic fusion algorithm dynamic library installed on the central node processor prototype. The updated wide-area optical scene image is then transmitted to the ground control software for display.

[0021] Step S3-3: Evaluate the fusion effect of the central node processor prototype and the wide-area optical scene dynamic fusion algorithm dynamic library on the image position.

[0022] Furthermore, in step S4, the optical target intelligent collaborative recognition specifically includes the following steps:

[0023] Step S4-1: The UAV cluster and the target drone fly according to the instructions of the ground control software;

[0024] Step S4-2: The central node drone subsystem uses the central node optical target intelligent collaborative recognition algorithm dynamic library to preliminarily identify the target drone in the scene, and sends a command to guide the edge node drone in the area where the target drone is located to fly to the designated location;

[0025] Step S4-3: After receiving the instruction from the central node processor prototype, the edge node UAV flies to the designated location. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library and feeds back the recognition result to the central node processor prototype.

[0026] Step S4-4: The central node UAV subsystem performs decision-level fusion recognition on the recognition results of the edge node UAV subsystem and transmits it to the ground control software; during the collaborative recognition process, the prototype recognition results are saved to the ground control software.

[0027] Furthermore, in step S5, the central node and the edge node collaborate to identify data confirmation and determination, specifically including the following steps:

[0028] Step S5-1: confirming the recognition effect of collaborative recognition from the saved collaborative recognition results, with the evaluation indicators including the target recognition confidence of the central node drone subsystem and the edge node drone subsystem and the collaborative recognition confidence;

[0029] Step S5-2: In an outdoor test environment, evaluate the effectiveness of the central node UAV subsystem in guiding the edge node UAV subsystem to perform collaborative identification and perform decision-level fusion on the identification results of the edge node UAV subsystem.

[0030] Furthermore, in step S6, the recognition accuracy statistics and evaluation specifically include the following steps:

[0031] Step S6-1: Manually confirm the saved recognition images in turn and calculate the average recognition accuracy of the target machine under typical examples:

[0032]

[0033] Among them, TP is the number of correctly identified drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, and are actually targets; FP is the number of falsely reported drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, but are not targets; FN is the number of falsely reported drones that are not identified as targets by the optical target intelligent collaborative recognition algorithm, but are actually targets.

[0034] Step S6-2: Determine whether the average recognition accuracy of the processing prototype in the field test under a typical example meets the actual requirements.

[0035] In addition, an embodiment of the present invention also provides a collaborative perception and recognition system based on drone clusters, which includes a central node drone subsystem, an edge node drone subsystem and ground control software.

[0036] Furthermore, the central node UAV subsystem includes a central node UAV, a central node processor prototype, a central sequence image optical scene construction algorithm dynamic library, a wide-area optical scene dynamic fusion algorithm dynamic library, and a central node optical target intelligent collaborative recognition algorithm dynamic library;

[0037] The central node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on a central node drone;

[0038] The dynamic library of the sequential image optical scene construction algorithm runs on the central node processor prototype to realize the sequential image optical scene construction. Its input is the image data captured by the edge node drone subsystem, and its output is the spliced ​​wide-area optical scene image;

[0039] The dynamic library of the wide-area optical scene dynamic fusion algorithm runs on the central node processor prototype, fusing the high-resolution local area image from the edge node drone subsystem with the low-resolution wide-area optical scene image to achieve local resolution enhancement of the wide-area optical scene image;

[0040] The central node optical target intelligent collaborative recognition algorithm dynamic library runs on the central node processor prototype, and the central node processor prototype uses the central node optical target intelligent collaborative recognition algorithm dynamic library to perform preliminary recognition and discovery of targets in the scene.

[0041] Furthermore, the edge node UAV subsystem includes an edge node UAV, an edge node processor prototype, and an edge node optical target intelligent collaborative recognition algorithm dynamic library;

[0042] The edge node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on an edge node drone;

[0043] The edge node optical target intelligent collaborative recognition dynamic library runs on the edge node processor prototype. The edge node UAV flies to the designated location after receiving the instruction of the central node processor prototype. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library.

[0044] Furthermore, the ground control software has interface display interaction, data communication, database management, image reception and real-time video decoding and display functions. The software design is based on a multi-threaded state machine mode, and the above functions are integrated into corresponding functional modules, namely, the interface interaction module, data communication module, database management module, image reception module and video decoding module. The result status and image refresh during software operation are jointly handled by the main thread and the interface interaction module.

[0045] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0046] 1. The collaborative perception and recognition method and system based on drone clusters proposed in the present invention fully utilizes the global scheduling and decision-making capabilities of the central node through the collaborative operation of the central node drone subsystem and the edge node drone subsystem, while utilizing the real-time computing capabilities of the edge node to achieve efficient coverage and target recognition of wide-area scenarios.

[0047] 2. The proposed central node is responsible for wide-area scene scanning, image stitching, and preliminary target recognition, while the edge nodes, following the central node's instructions, perform high-resolution image acquisition and target confirmation in specific areas. This collaborative model not only improves the system's overall detection efficiency but also significantly enhances the accuracy and real-time nature of target recognition, resolving the computational load imbalance and information redundancy issues inherent in traditional centralized or distributed architectures.

[0048] 3. This invention utilizes a dynamic library of algorithms for constructing sequential optical scenes and a dynamic library of algorithms for fusion of wide-area optical scenes. This allows the central node to fuse high-resolution local images from edge nodes with low-resolution wide-area optical scene images, enhancing the local resolution of wide-area optical scene images. This technical solution effectively addresses the issue of insufficient wide-area scene image resolution in traditional methods, while avoiding the loss of edge information caused by block processing. This significantly improves the overall quality of wide-area scene images and their ability to capture small objects.

[0049] 4. This invention utilizes a dynamic library of optical target intelligent collaborative recognition algorithms for both the central node and the edge node, enabling preliminary target recognition and further confirmation. This two-stage recognition mechanism not only improves target recognition accuracy but also further reduces the probability of false and missed detections through decision-level fusion, providing reliable support for real-time monitoring and decision-making in complex environments.

[0050] 5. The ground control system proposed in this invention is designed based on a multi-threaded state machine model, integrating functional modules such as interface display interaction, data communication, database management, image reception, and real-time video decoding and display. This design not only improves system operational efficiency and real-time data processing, but also enhances system scalability and maintainability through modular design, ensuring the long-term stable operation of the drone swarm system.

[0051] 6. Both the central node processor prototype and the edge node processor prototype utilize the NVIDIA Jetson NX core board processor, combined with peripheral interface circuit boards, solid-state drives, and cooling fans, providing high-performance, low-power computing support. This hardware design not only meets the real-time computing needs of the drone's onboard terminal, but also extends the drone's flight time through its low-power design, providing hardware support for long-term detection in wide-area scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a collaborative perception and recognition method based on drone clusters proposed by the present invention.

[0053] Figure 2 A flowchart of constructing a wide-area optical scene according to an embodiment of the present invention.

[0054] Figure 3 Schematic diagram of drone cluster control constructed from a sequence of image optical scenes provided by an embodiment of the present invention.

[0055] Figure 4 This is a diagram of the wide-area optical scene construction interface of the ground control software provided in an embodiment of the present invention.

[0056] Figure 5 This is a diagram of the wide-area optical scene dynamic fusion interface of the ground control software provided in an embodiment of the present invention.

[0057] Figure 6 This is a flowchart of the intelligent collaborative recognition of optical targets provided by an embodiment of the present invention.

[0058] Figure 7 Schematic diagram of the control of a drone cluster and a target drone cluster for intelligent collaborative identification of optical targets provided by an embodiment of the present invention.

[0059] Figure 8 An optical target intelligent collaborative recognition interface for ground control software provided by an embodiment of the present invention.

[0060] Figure 9 This is a structural diagram of the collaborative perception and recognition system based on drone clusters proposed in this invention. DETAILED DESCRIPTION

[0061] The following further describes the specific embodiments of the present invention in conjunction with the accompanying drawings, clearly and completely describing the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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.

[0062] Combine Figure 1 , the present invention proposes a collaborative perception and recognition method based on drone clusters, which specifically includes the following steps:

[0063] Step S1: Start the ground control software and load the UAV route path file;

[0064] Step S2: The central node UAV subsystem receives the image data captured by the edge node UAV subsystem, constructs a sequence image optical scene using the algorithm dynamic library carried by the central node UAV subsystem, and evaluates the effect of the wide-area optical scene construction;

[0065] Step S3: The edge node UAV subsystem collects high-resolution image data and sends it to the central node UAV subsystem. The central node UAV subsystem uses the algorithm dynamic library to perform wide-area optical scene dynamic fusion and evaluate the wide-area optical scene dynamic fusion effect.

[0066] Step S4: The central node UAV subsystem preliminarily identifies the target drone and guides the edge node UAV subsystem. The edge node UAV subsystem feeds back the high-resolution recognition results to the central node UAV subsystem, which then performs decision-level fusion recognition to achieve intelligent collaborative recognition of optical targets.

[0067] Step S5: Collaborative identification data confirmation and determination;

[0068] Step S6: Recognition accuracy statistics and evaluation.

[0069] Further, combined Figure 2 In step S2, the construction and effect evaluation of the optical scene of the sequence image specifically include the following steps:

[0070] Step S2-1: The center node UAV and the edge node UAV fly according to the cluster control instructions of the ground control software and perform field detection to collect image data; five node UAVs (1 center and 4 edge) fly along parallel straight lines to shoot, with a flight altitude of 100m and a flight speed of 5m / s. Each node shoots a frame of image every 2s; the UAV flight path is as follows: Figure 3 As shown in the figure, when the scene is constructed, the diagonal field of view of each node camera is 82.9 degrees. Considering the field of view overlap rate, the distance between adjacent nodes is about 35m, and the flight path distance is about 130m.

[0071] Step S2-2: The central node UAV processor prototype receives the image data captured by the edge node UAV subsystem, and realizes the sequence image optical scene construction through the sequence image optical scene construction algorithm dynamic library installed on the central node processor prototype, and transmits the constructed optical scene image to the ground control software for display. The wide-area optical scene construction interface of the ground control software is shown in the figure below. Figure 4 As shown;

[0072] Step S2-3: Evaluate the wide-area optical scene image effect constructed by the central node processor prototype and the dynamic library of the sequential image optical scene construction algorithm it carries. The evaluation indicators include visual distortion and splicing misalignment.

[0073] Furthermore, in step S3, the dynamic fusion and effect evaluation of the wide-area optical scene specifically includes the following steps:

[0074] Step S3-1: The edge node drone improves the field of view image resolution by zooming and collects high-resolution image data;

[0075] Step S3-2: The edge node UAV subsystem sends the high-resolution image data to the central node UAV subsystem, and the high-resolution image is dynamically fused to the corresponding area of ​​the wide-area optical scene through the dynamic library of the wide-area optical scene dynamic fusion algorithm installed on the central node processor prototype, and the updated wide-area optical scene image is transmitted to the ground control software for display; wherein, the frame rate of the image dynamic fusion is about 2Hz; the wide-area optical scene dynamic fusion interface of the ground control software is as shown in the figure Figure 5 As shown;

[0076] Step S3-3: Evaluate the fusion effect of the central node processor prototype and the wide-area optical scene dynamic fusion algorithm dynamic library on the image position.

[0077] Further, combined Figure 6 In step S4, the optical target intelligent collaborative recognition specifically includes the following steps:

[0078] Step S4-1: The UAV cluster and the target drone fly according to the instructions of the ground control software; Figure 7 As shown, five node drones (1 main and 4 edge drones) took off to an altitude of about 190m. The main node camera's field of view (DFOV) was 47°, and the edge node camera's field of view (DFOV) was 12°. The main node's field of view basically covered the edge nodes. Five target drones took off (4 small rotorcraft and 1 fixed-wing). The target drones flew about 30-150m above the ground near the main node's field of view.

[0079] Step S4-2: The central node drone subsystem uses the central node optical target intelligent collaborative recognition algorithm dynamic library to preliminarily identify the target drone in the scene, and sends a command to guide the edge node drone in the area where the target drone is located to fly to the designated location;

[0080] Step S4-3: After receiving the instruction from the central node processor prototype, the edge node UAV flies to the designated location. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library and feeds back the recognition result to the central node processor prototype.

[0081] Step S4-4: The central node UAV subsystem performs decision-level fusion recognition on the recognition results of the edge node UAV subsystem and transmits it to the ground control software. The optical target intelligent collaborative recognition interface of the ground control software is as follows: Figure 8 As shown; during the collaborative identification process, the prototype identification results are saved to the ground control software.

[0082] Furthermore, in step S5, the central node and the edge node collaborate to identify data confirmation and determination, specifically including the following steps:

[0083] Step S5-1: confirming the recognition effect of collaborative recognition from the saved collaborative recognition results, with the evaluation indicators including the target recognition confidence of the central node drone subsystem and the edge node drone subsystem and the collaborative recognition confidence;

[0084] Step S5-2: In an outdoor test environment, evaluate the effectiveness of the central node UAV subsystem in guiding the edge node UAV subsystem to perform collaborative identification and perform decision-level fusion on the identification results of the edge node UAV subsystem.

[0085] Furthermore, in step S6, the recognition accuracy statistics and evaluation specifically include the following steps:

[0086] Step S6-1: Manually confirm the saved recognition images in turn and calculate the average recognition accuracy of the target machine under typical examples:

[0087]

[0088] Among them, TP is the number of correctly identified drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, and are actually targets; FP is the number of falsely reported drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, but are not targets; FN is the number of falsely reported drones that are not identified as targets by the optical target intelligent collaborative recognition algorithm, but are actually targets.

[0089] Step S6-2: Determine whether the average recognition accuracy of the processing prototype in the field test under a typical example meets the actual requirements.

[0090] In addition, an embodiment of the present invention also provides a collaborative perception and recognition system based on drone clusters, which includes a central node drone subsystem, an edge node drone subsystem and ground control software.

[0091] Further, combined Figure 9 The central node UAV subsystem includes a central node UAV, a central node processor prototype, a central sequence image optical scene construction algorithm dynamic library, a wide-area optical scene dynamic fusion algorithm dynamic library, and a central node optical target intelligent collaborative recognition algorithm dynamic library;

[0092] The central node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on a central node drone;

[0093] The dynamic library of the sequential image optical scene construction algorithm runs on the central node processor prototype to realize the sequential image optical scene construction. Its input is the image data captured by the edge node drone subsystem, and its output is the spliced ​​wide-area optical scene image;

[0094] The dynamic library of the wide-area optical scene dynamic fusion algorithm runs on the central node processor prototype, fusing the high-resolution local area image from the edge node drone subsystem with the low-resolution wide-area optical scene image to achieve local resolution enhancement of the wide-area optical scene image;

[0095] The central node optical target intelligent collaborative recognition algorithm dynamic library runs on the central node processor prototype, and the central node processor prototype uses the central node optical target intelligent collaborative recognition algorithm dynamic library to perform preliminary recognition and discovery of targets in the scene.

[0096] Furthermore, the edge node UAV subsystem includes an edge node UAV, an edge node processor prototype, and an edge node optical target intelligent collaborative recognition algorithm dynamic library;

[0097] The edge node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on an edge node drone;

[0098] The edge node optical target intelligent collaborative recognition dynamic library runs on the edge node processor prototype. The edge node UAV flies to the designated location after receiving the instruction of the central node processor prototype. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library.

[0099] Furthermore, the ground control software has interface display interaction, data communication, database management, image reception and real-time video decoding and display functions. The software design is based on a multi-threaded state machine mode, and the above functions are integrated into corresponding functional modules, namely, the interface interaction module, data communication module, database management module, image reception module and video decoding module. The result status and image refresh during software operation are jointly handled by the main thread and the interface interaction module.

[0100] The above implementation scheme is only the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical scheme in accordance with the technical idea proposed by the present invention fall within the protection scope of the present invention.

Claims

1. A collaborative perception and recognition method based on drone clusters, characterized in that: The specific steps include: Step S1: Start the ground control software and load the UAV route path file; Step S2: The central node UAV subsystem receives the image data captured by the edge node UAV subsystem, constructs a sequence image optical scene using the algorithm dynamic library carried by the central node UAV subsystem, and evaluates the effect of the wide-area optical scene construction; Step S3: The edge node UAV subsystem collects high-resolution image data and sends it to the central node UAV subsystem. The central node UAV subsystem uses the algorithm dynamic library to perform wide-area optical scene dynamic fusion and evaluate the wide-area optical scene dynamic fusion effect. Step S4: The central node UAV subsystem preliminarily identifies the target drone and guides the edge node UAV subsystem. The edge node UAV subsystem feeds back the high-resolution recognition results to the central node UAV subsystem, which then performs decision-level fusion recognition to achieve intelligent collaborative recognition of optical targets. Step S5: Collaborative identification data confirmation and determination; Step S6: Recognition accuracy statistics and evaluation.

2. The collaborative perception and recognition method based on drone swarms according to claim 1, characterized in that: In step S2, the construction and effect evaluation of the optical scene of the sequence of images specifically include the following steps: Step S2-1: The central node UAV and the edge node UAV fly according to the cluster control instructions of the ground control software and perform field detection to collect image data; Step S2-2: The central node UAV processor prototype receives image data captured by the edge node UAV subsystem, constructs a sequence image optical scene using the sequence image optical scene construction algorithm dynamic library installed on the central node processor prototype, and transmits the constructed optical scene image to the ground control software for display; Step S2-3: Evaluate the wide-area optical scene image effect constructed by the central node processor prototype and the dynamic library of the sequential image optical scene construction algorithm it carries. The evaluation indicators include visual distortion and splicing misalignment.

3. The collaborative perception and recognition method based on drone swarms according to claim 1, characterized in that: In step S3, the dynamic fusion and effect evaluation of the wide-area optical scene specifically include the following steps: Step S3-1: The edge node drone improves the field of view image resolution by zooming and collects high-resolution image data; Step S3-2: The edge node UAV subsystem sends the high-resolution image data to the central node UAV subsystem. The high-resolution image is dynamically fused to the corresponding area of ​​the wide-area optical scene using the wide-area optical scene dynamic fusion algorithm dynamic library installed on the central node processor prototype. The updated wide-area optical scene image is then transmitted to the ground control software for display. Step S3-3: Evaluate the fusion effect of the central node processor prototype and the wide-area optical scene dynamic fusion algorithm dynamic library on the image position.

4. The collaborative perception and recognition method based on drone swarms according to claim 1, characterized in that: In step S4, the optical target intelligent collaborative recognition specifically includes the following steps: Step S4-1: The UAV cluster and the target drone fly according to the instructions of the ground control software; Step S4-2: The central node drone subsystem uses the central node optical target intelligent collaborative recognition algorithm dynamic library to preliminarily identify the target drone in the scene, and sends a command to guide the edge node drone in the area where the target drone is located to fly to the designated location; Step S4-3: After receiving the instruction from the central node processor prototype, the edge node UAV flies to the designated location. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library and feeds back the recognition result to the central node processor prototype. Step S4-4: The central node UAV subsystem performs decision-level fusion recognition on the recognition results of the edge node UAV subsystem and transmits it to the ground control software; during the collaborative recognition process, the prototype recognition results are saved to the ground control software.

5. The collaborative perception and recognition method based on drone swarms according to claim 1, characterized in that: In step S5, the central node and the edge node collaborate to identify data confirmation and determination, specifically including the following steps: Step S5-1: confirming the recognition effect of collaborative recognition from the saved collaborative recognition results, with the evaluation indicators including the target recognition confidence of the central node drone subsystem and the edge node drone subsystem and the collaborative recognition confidence; Step S5-2: In an outdoor test environment, evaluate the effectiveness of the central node UAV subsystem in guiding the edge node UAV subsystem to perform collaborative identification and perform decision-level fusion on the identification results of the edge node UAV subsystem.

6. The collaborative perception and recognition method based on drone swarms according to claim 1, characterized in that: In step S6, the recognition accuracy statistics and evaluation specifically include the following steps: Step S6-1: Manually confirm the saved recognition images in turn and calculate the average recognition accuracy of the target machine under typical examples: Among them, TP is the number of correctly identified drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, and are actually targets; FP is the number of falsely reported drones that are identified as targets by the optical target intelligent collaborative recognition algorithm, but are not targets; FN is the number of falsely reported drones that are not identified as targets by the optical target intelligent collaborative recognition algorithm, but are actually targets. Step S6-2: Determine whether the average recognition accuracy of the processing prototype in the field test under a typical example meets the actual requirements.

7. A collaborative perception and recognition system based on drone clusters, characterized by: The collaborative perception and recognition system based on drone clusters is applied to a collaborative perception and recognition method based on drone clusters as described in any one of claims 1-6. The collaborative perception and recognition system based on drone clusters includes a central node drone subsystem, an edge node drone subsystem and ground control software.

8. The collaborative perception and recognition system based on drone swarms according to claim 7, characterized in that: The central node UAV subsystem includes a central node UAV, a central node processor prototype, a dynamic library of sequential image optical scene construction algorithms, a dynamic library of wide-area optical scene dynamic fusion algorithms, and a dynamic library of central node optical target intelligent collaborative recognition algorithms. The central node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on a central node drone; The dynamic library of the sequential image optical scene construction algorithm runs on the central node processor prototype to realize the sequential image optical scene construction. Its input is the image data captured by the edge node drone subsystem, and its output is the spliced ​​wide-area optical scene image; The dynamic library of the wide-area optical scene dynamic fusion algorithm runs on the central node processor prototype, fusing the high-resolution local area image from the edge node drone subsystem with the low-resolution wide-area optical scene image to achieve local resolution enhancement of the wide-area optical scene image; The central node optical target intelligent collaborative recognition algorithm dynamic library runs on the central node processor prototype, and the central node processor prototype uses the central node optical target intelligent collaborative recognition algorithm dynamic library to perform preliminary recognition and discovery of targets in the scene.

9. The collaborative perception and recognition system based on drone swarms according to claim 7, characterized in that: The edge node UAV subsystem includes an edge node UAV, an edge node processor prototype, and an edge node optical target intelligent collaborative recognition algorithm dynamic library; The edge node processor prototype consists of an NVIDIA Jetson NX core board processor, a peripheral interface circuit board, a solid-state drive, a cooling fan, and a casing, and is carried on an edge node drone; The edge node optical target intelligent collaborative recognition dynamic library runs on the edge node processor prototype. The edge node UAV flies to the designated location after receiving the instruction of the central node processor prototype. The edge node processor prototype further identifies and confirms the target through the edge node optical target intelligent collaborative recognition dynamic library.

10. The collaborative perception and recognition system based on drone swarms according to claim 7, characterized in that: The ground control software has interface display interaction, data communication, database management, image reception and real-time video decoding and display functions. The software design is based on a multi-threaded state machine mode, and the above functions are integrated into corresponding functional modules, namely the interface interaction module, data communication module, database management module, image reception module and video decoding module. The result status and image refresh during software operation are jointly handled by the main thread and the interface interaction module.

Images