Perception and control integrated method for unmanned aerial vehicle to intercept low-slow small target, electronic equipment and medium

Abstract

The invention discloses a perception and control integration method for an unmanned aerial vehicle to intercept a low-slow small target, electronic equipment and a medium. The method comprises the following steps: constructing a control scheme according to an unmanned aerial vehicle perception-control integrated interception system; constructing a target detection network BiGNet; and designing a tracking and intercepting controller for the low-slow small target. According to the method, target detection and control depth are coupled by using a model prediction control method, efficient identification and stable tracking of the unmanned aerial vehicle on the low-slow small target under a complex background are realized, meanwhile, the safety and stability of the formation are kept, and the method has practicability in practical application of intercepting and countering low-slow small unmanned aerial vehicle tasks.

Description

Technical Field

[0001] The present invention relates to the field of multi-agent decision control, and more specifically, to an integrated perception and control method, electronic equipment and medium for unmanned aerial vehicles to intercept low, slow and small targets. Background Art

[0002] With the rapid development of the low-altitude economy, drone technology has been widely used in many fields, such as logistics, agricultural monitoring, and security. However, the widespread use of low, slow, and small target (LSA) drones has brought huge challenges to traditional drone detection and countermeasure systems. These low, slow, and small drones are usually small in size, slow in flight speed, and have complex flight trajectories. They often fly in complex environments such as cities and airports, which makes it difficult for existing detection technologies to effectively identify their existence. Especially in the presence of a large amount of interference and occlusion, the accuracy and real-time performance of target detection often cannot meet the requirements.

[0003] Traditional UAV detection and countermeasure methods mostly adopt a separate architecture, that is, the perception and control modules operate independently, which cannot achieve deep information fusion, resulting in a delayed response of the system when dealing with rapidly changing dynamic targets. Especially in target tracking tasks involving multiple UAVs working together or in complex environments, traditional methods often have problems such as low target recognition accuracy, unstable target tracking, and poor flight formation safety. At present, countermeasures against low, slow, and small targets mostly rely on traditional sensors such as radar and infrared detection, but these methods have limited adaptability and robustness in complex backgrounds, especially for the detection and identification of small UAVs, which presents great technical difficulties. At the same time, traditional control strategies usually ignore the impact of perception information on flight paths and mission execution, resulting in the inability to achieve efficient dynamic target tracking and countermeasures.

[0004] Therefore, it is necessary to develop an integrated perception and control method, electronic equipment and medium for UAVs to intercept low, slow and small targets.

[0005] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0006] The present invention proposes an integrated perception and control method, electronic device and medium for unmanned aerial vehicles to intercept low, slow and small targets. The method can use the model predictive control method to deeply couple target detection and control, thereby realizing the efficient recognition and stable tracking of low, slow and small targets by unmanned aerial vehicles in complex backgrounds, while maintaining the safety and stability of the formation. The method is practical in solving the actual application of the task of intercepting and countering low, slow and small unmanned aerial vehicles.

[0007] In a first aspect, an embodiment of the present disclosure provides a perception and control integrated method for a drone to intercept a low, slow, and small target, including:

[0008] Build a control solution based on the drone perception-control integrated interception system;

[0009] Build the target detection network BiGNet;

[0010] Design a tracking and interception controller for low, slow and small targets.

[0011] Preferably, the control scheme includes:

[0012] Obtain original image information and identify the target drone through an aerial target detector;

[0013] The perception information is integrated into the tracking controller for low, slow and small targets, and combined with the tracking disturbance compensator, a model predictive controller is constructed to control the UAV.

[0014] Preferably, constructing the target detection network BiGNet includes:

[0015] A bidirectional feature pyramid network is used as the feature fusion mechanism to determine the feature fusion process;

[0016] For the fusion process of the middle layer features, the features of each layer are combined through a weighted mechanism to determine the fusion features;

[0017] The GhostConv module is introduced in BiGNet, which uses half of the features for convolution operations and concatenates the generated features with the original features to create redundant features.

[0018] Preferably, the feature fusion process is:

[0019]

[0020] Among them, w i ≥0 is the learned weight, I i is the input feature map, and ε is a stable numerical coefficient.

[0021] Preferably, the fusion features are:

[0022] Among them, P i Td is the intermediate feature of the i-th layer, P i Out is the output feature of the i-th layer, Conv is a depth-wise separable convolution operation, and Resize is an upsampling or downsampling operation.

[0023] Preferably, the tracking and interception controller designed for low, slow and small targets includes:

[0024] Visual perception information is introduced into the MPC optimization framework, and the cost functions of target tracking and disturbance compensation are added for joint optimization to obtain a tracking controller for low, slow and small targets.

[0025] Preferably, the tracking controller for low, slow and small targets is:

[0026]

[0027] in,

[0028] Preferably, it also includes:

[0029] By introducing disturbance conditions, the anti-disturbance ability and tracking stability of the tracking controller for low, slow and small targets in complex environments are verified.

[0030] In a second aspect, an embodiment of the present disclosure further provides an electronic device, the electronic device comprising:

[0031] A memory storing executable instructions;

[0032] A processor runs the executable instructions in the memory to implement the perception and control integration method for unmanned aerial vehicle interception of low, slow and small targets.

[0033] In a third aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the integrated perception and control method for drones to intercept low, slow and small targets is implemented.

[0034] Its beneficial effects are:

[0035] (1) Based on the optimization algorithm of lightweight convolution and feature fusion, the detection accuracy and speed of low, slow and small targets in complex backgrounds are significantly improved. The visual detection results can be integrated into the MPC controller, and a UAV tracking control framework with integrated perception and control is designed to achieve efficient target tracking and formation stability.

[0036] (2) In order to deal with the situation where dense formations are disturbed by aerodynamic forces, Gaussian process regression is used to model the aerodynamic forces between UAVs, and real-time error compensation is implemented in the control algorithm, which improves the stability of the formation when intercepting low, slow and small targets;

[0037] (3) In solving the problem of intercepting low, slow and small targets in complex environments, the present invention has significant advantages such as efficient target detection, deep coupling of perception and control, real-time compensation of formation errors and comprehensive data support. It not only improves the detection and tracking capabilities of low, slow and small targets, but also ensures the stability and safety of UAV formations in complex environments, and has good practicality.

[0038] The methods and apparatus of the present invention have other features and advantages that will be apparent from, or will be described in detail in, the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0040] Figure 1 A flowchart showing the steps of a perception-control integration method for a drone to intercept a low, slow, and small target according to an embodiment of the present invention is shown.

[0041] Figure 2 A schematic diagram of an architecture diagram of a UAV perception-control integrated interception system according to an embodiment of the present invention is shown.

[0042] Figure 3 A schematic diagram of a low-speed, small target detector network structure according to an embodiment of the present invention is shown.

[0043] Figure 4 A schematic diagram showing the stability results of low, slow and small target tracking according to an embodiment of the present invention.

[0044] Figure 5 A schematic diagram showing the tracking accuracy results of a slow, small target according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0046] To facilitate understanding of the solutions and effects of the embodiments of the present invention, three specific application examples are given below. Those skilled in the art should understand that the examples are only for facilitating understanding of the present invention, and any specific details thereof are not intended to limit the present invention in any way.

[0047] Example 1

[0048] Figure 1 A flowchart showing the steps of a perception-control integration method for a drone to intercept a low, slow, and small target according to an embodiment of the present invention is shown.

[0049] like Figure 1 As shown, the perception and control integration method for UAV to intercept low, slow and small targets includes:

[0050] Step 101, constructing a control scheme according to the UAV perception-control integrated interception system;

[0051] Step 102, constructing a target detection network BiGNet;

[0052] Step 103, design a tracking and interception controller for low, slow and small targets.

[0053] In one example, the control scheme includes:

[0054] Obtain original image information and identify the target drone through an aerial target detector;

[0055] The perception information is integrated into the tracking controller for low, slow and small targets, and combined with the tracking disturbance compensator, a model predictive controller is constructed to control the UAV.

[0056] In one example, building a target detection network BiGNet includes:

[0057] A bidirectional feature pyramid network is used as the feature fusion mechanism to determine the feature fusion process;

[0058] For the fusion process of the middle layer features, the features of each layer are combined through a weighted mechanism to determine the fusion features;

[0059] The GhostConv module is introduced in BiGNet, which uses half of the features for convolution operations and concatenates the generated features with the original features to create redundant features.

[0060] In one example, the feature fusion process is:

[0061]

[0062] Among them, w i ≥0 is the learned weight, I i is the input feature map, and ε is a stable numerical coefficient.

[0063] In one example, the fused features are:

[0064] Among them, Pi Td is the intermediate feature of the i-th layer, P i Out is the output feature of the i-th layer, Conv is a depth-wise separable convolution operation, and Resize is an upsampling or downsampling operation.

[0065] In one example, designing a tracking and interception controller for a low, slow, and small target includes:

[0066] Visual perception information is introduced into the MPC optimization framework, and the cost functions of target tracking and disturbance compensation are added for joint optimization to obtain a tracking controller for low, slow and small targets.

[0067] In one example, the tracking controller for a low, slow, small target is:

[0068]

[0069] stx k∣k =x init

[0070] x k+i+1∣k =f pred (x k+i∣k ,u k+i∣k ,δ p,k+i )

[0071]

[0072] 0≤α i ≤α max

[0073] in,

[0074] In one example, it also includes:

[0075] By introducing disturbance conditions, the anti-disturbance ability and tracking stability of the tracking controller for low, slow and small targets in complex environments are verified.

[0076] Specifically, the present invention proposes an integrated perception and control method for the identification and tracking of low, slow and small targets in complex urban environments, which mainly solves the problems of low detection accuracy of low, slow and small targets, unstable tracking, and difficulty in collaborative control of multiple UAVs. By tightly coupling the target detection algorithm with the UAV formation control algorithm, efficient target identification and stable tracking in complex backgrounds are achieved. The present invention uses lightweight convolution and efficient feature fusion technology to optimize target detection performance, designs an integrated perception and control framework in combination with model predictive control, and realizes formation error compensation through Gaussian process regression method, which effectively improves the robustness of target tracking and the stability of the formation. In addition, through a self-developed simulation platform, a high-quality anti-UAV target dataset covering diverse backgrounds of cities and airports was constructed, providing solid data support for this technology.

[0077] Figure 2 A schematic diagram of an architecture diagram of a UAV perception-control integrated interception system according to an embodiment of the present invention is shown.

[0078] like Figure 2 As shown in the figure, the original image information is obtained from the drone's own camera, the target drone is identified through the aerial target detector, and after a series of processing such as coordinate transformation, the target position and other information are obtained; then the system integrates the perception information into the tracking controller for low, slow and small targets, combines with the tracking disturbance compensator, and builds a model predictive controller to control the drone, which can improve the accuracy of tracking low, slow and small targets while maintaining stable tracking of the target.

[0079] Figure 3 A schematic diagram of a low-speed, small target detector network structure according to an embodiment of the present invention is shown.

[0080] Build a low-speed small object detector, such as Figure 3 As shown in the figure, BiGNet is a lightweight and efficient network model based on YOLOv10.

[0081] In BiGNet, a bidirectional feature pyramid network (BiFPN) is used as the core feature fusion mechanism, aiming to effectively fuse multi-scale features input from the backbone network. Specifically, the BiFPN structure has the following important improvements for the task. First, the network structure is simplified, redundant feature fusion nodes are removed, and only valid input edges are retained, which reduces the computational complexity without reducing performance. Secondly, by adding additional feature branches between the original input and output nodes of the same layer, the richness of feature fusion is improved, thereby enhancing the expressiveness of the network. Finally, due to the difference in resolution of feature maps of different scales, BiFPN balances the influence of each feature layer through a trainable weighting mechanism to achieve deep fusion of features. The BiFPN feature fusion process uses a dynamic weighting mechanism to adjust the contribution of different feature maps. Considering the training efficiency and performance, a fast normalization fusion method is selected. The formula is as follows:

[0082]

[0083] Among them, w i ≥0 is the learned weight, I i is the input feature map, and the stable numerical coefficient is ε=0.0001.

[0084] For the fusion process of the intermediate layer features, the features of each layer are combined through a weighted mechanism. The specific formula of the two fused features is:

[0085] Among them, P i Td is the intermediate feature of the i-th layer, P i Out is the output feature of the i-th layer, Conv is a depth-separable convolution operation, and Resize is an upsampling or downsampling operation. Through these improvements, BiFPN can not only effectively capture the location information and semantic information in complex scenes, but also ensure more accurate fusion between features of different scales. The BiFPN network structure demonstrates the overall picture and efficiency of its feature fusion. By optimizing the feature fusion strategy, BiFPN in BiGNet significantly improves YOLOv10's ability to detect illegal low-speed and slow small drones in complex urban environments, and achieves accurate and efficient target recognition.

[0086] The GhostConv module is introduced in BiGNet to achieve efficient feature extraction and redundant feature generation. Specifically, GhostConv first uses half of the features for convolution operations, and then concatenates the generated features with the original features to create redundant features. This strategy not only reduces the number of model parameters and computational complexity, but also effectively improves detection accuracy.

[0087] Combining the GhostConv module with the BiFPN network structure, an efficient target detection network BiGNet is proposed. Through this design, BiGNet significantly reduces the number of network parameters while maintaining high detection accuracy, ensuring its applicability to resource-constrained edge devices. The flexibility and efficiency of the network enable it to maintain excellent detection capabilities in complex scenarios.

[0088] Design a tracking and interception controller for low, slow and small targets. Introduce visual perception information into the MPC optimization framework, and further constrain and optimize the control process by combining the visual perception data of the drone. According to the classic pinhole camera model, the coordinates of the projection point of the center point of the low, slow and small target on the image plane can be obtained as [u c ,v c ], define the coordinates of the center point of the image as [u o ,v o ], we can get the distance r between the two on the image:

[0089] r=||(u c ,v c )-(u o ,v o )||2=||u c -u o ,v c -v o ||2 where ||·||2 represents the Euclidean norm, which is used to calculate the distance between the target drone projection point and the image center. A new position cost variable is introduced for MPC

[0090] The variable d represents the distance of a low, slow, small target in the body coordinate system. T Indicates the set value for the distance that the counter-drone should maintain from the low, slow and small target. T The specific value needs to be determined according to actual needs.

[0091] Gaussian process is a random process, and any finite number of random variables follow a joint probability distribution. Using this feature, the Gaussian process regression method is used to fit the impact of the downwash effect caused by aerodynamic disturbances on the state of the drone, and the residual model f is established. pred =f real -f nom , where f real represents the true model, f nom The position difference δ between the four-rotor drones is used as the model. p As input, the output is the influence of the mutual aerodynamic force on the acceleration of the body Assume that the sample set {f(x)} satisfies:

[0092] Where μ(x) is the prior mean function and κ(x,x') is the covariance function. The mean function μ(x) = 0, and the covariance function is the square exponential covariance function:

[0093]

[0094] is the data variance, L d is a positive diagonal matrix. Assume that the noise is Gaussian noise and obeys N(0,σ 2 ), and then, we can get the Gaussian regression process prediction part:

[0095] Data is collected in real time during the flight of the UAV. For the quadrotor i, at each sampling point t k-1 If it is determined that there is only one neighbor, then record the state x k-1 , control input u k-1 and the position difference δ between the adjacent nodes at this time p,k-1 When the next sampling time t is reached k When k , and update the data set online according to the following formula to perform real-time disturbance correction:

[0096]

[0097] make Define two thresholds and when or Update the dataset

[0098] In order to strike a balance between accurate and stable target tracking, the cost functions of target tracking and disturbance compensation are added to the classic MPC controller for joint optimization. During the optimization process, the tracking cost is mainly reflected in the deviation between the target and the center of the camera's field of view and the measured depth. This part of the cost reflects the tracking accuracy of the UAV on the target, ensuring that the target always remains in the center of the camera's field of view and maintains an appropriate distance. Disturbance compensation reflects the stability constraints added by the UAV to the aerodynamic effects between aircraft during target tracking. By introducing the parameter α∈[0,α max ], MPC performs a weighted combination of these cost functions and dynamically adjusts the control input, so that it can accurately approach the target while maintaining stable tracking of the target drone. When α is close to 0, the system pays more attention to the target position in the visual perception information, that is, tracking accuracy is prioritized. On the contrary, when α is close to α maxThe system pays more attention to its own flight stability, that is, disturbance compensation has a higher weight. This dynamic trade-off mechanism enables the system to flexibly adjust the control input according to the flight state. When the UAV is close to the target, the accuracy of visual perception information is higher, so more priority is given to accurate tracking of the target; when the UAV is subject to strong inter-machine aerodynamic effects, it is more necessary to prioritize maintaining the stability of the tracked target. Finally, combining visual perception information and disturbance compensation mechanism, the tracking controller for low, slow and small targets is in the form of:

[0099]

[0100] stx k∣k =x init

[0101] x k+i+1∣k =f pred (x k+i∣k ,u k+i∣k ,δ p,k+i )

[0102]

[0103] 0≤α i ≤α max

[0104] in, The controller performs a weighted combination of these two cost functions at each moment, thereby ensuring the target tracking accuracy while maintaining the tracking stability of the UAV.

[0105] In order to verify the tracking performance of the model predictive controller (tracking controller for low, slow and small targets) with integrated visual perception and tracking control for low, slow and small targets in complex environments, a series of experiments were designed to compare the performance of the classic MPC without sensor-control integration and the tracking controller for low, slow and small targets with sensor-control integration. The experiment consists of two main parts: first, by introducing disturbance conditions, the anti-disturbance ability and tracking stability of the tracking controller for low, slow and small targets in complex environments are verified; second, the differences in target visibility between the two controllers are compared. The core of the experiment is to evaluate how the tracking controller for low, slow and small targets maintains the perception stability of the target in a disturbed environment and improves the accuracy of target tracking through visual perception information.

[0106] Figure 4 A schematic diagram showing the stability results of low, slow and small target tracking according to an embodiment of the present invention.

[0107] In the anti-disturbance and tracking and interception stability experiment, three quad-rotor drones were designed to track and intercept the target in a coordinated manner. Their flight paths were the same, but their heights differed by 0.5 meters. The experiment simulated the impact of downwash disturbances between drones in reality, and focused on examining the tracking stability of the drone group under disturbance conditions. The results are as follows: Figure 4 As shown, Figure 4 The displacement changes of the three drones in the downwash influence area and their deviations from the reference trajectory are shown. It can be seen that within the first 40m, the three drones are significantly affected by the downwash, and the drone at the bottom is the most seriously affected. After updating the database online and compensating the influence of the downwash through the Gaussian regression process, the formation configuration of the three drones gradually stabilized after 25m. In the y-axis direction, the three drones all fluctuated to varying degrees after entering the downwash influence area, but the final convergence effect was good. In the z-axis direction, the heights of the three drones were also disturbed to varying degrees. The fluctuation of drone 2 was the most obvious, with an average error of 0.0450m, while the error of drone 3 was relatively small at 0.0168m. Overall, although the downwash interfered with the flight of the drones, the three drones were able to quickly return to stability. The experimental results show that when the drone group performs the tracking and interception mission of a low, slow and small target, facing the situation where the aerodynamic force between the aircraft caused by the dense formation affects the target tracking and interception, the tracking controller can quickly restore its own stability to ensure stable detection and tracking of the target.

[0108] In the target tracking, interception and visibility experiments, the classic MPC assumes that the true position of the target can be directly obtained, while the tracking controller for low, slow and small targets uses the visual perception information obtained by the low, slow and small target detector to control the drone to track the target object. Experiments were conducted in three different urban environments, namely airports, water surfaces and forest backgrounds. These environments were selected to test the tracking effect of the tracking controller for low, slow and small targets on the target in scenes with large differences in visual features.

[0109] In order to quantitatively evaluate the tracking effect, a similarity score S is introduced to indicate the visibility of the target in the drone camera field of view. The specific formula of the similarity score is as follows:

[0110] S1=S box

[0111]

[0112] S=(S1·S2) 2

[0113] S boxrepresents the area of ​​the normalized target detection box to represent the effect of the controller in tracking low, slow and small targets. The value of h represents the normalization by the length of the image diagonal, where w and l are the width and height of the image, respectively. The experimental results of three different environments are shown in Table 1. The low, slow and small target tracking controller is significantly better than the classic MPC in all environments, and the target visibility is improved by 20% on average. The experimental results show that the integrated sensing and control tracking controller exhibits stronger target tracking ability and anti-disturbance performance in complex environments. By coupling visual perception information, it can effectively improve the visibility of the target under different backgrounds and ensure the stability of continuous tracking of the target under external disturbances.

[0114] Table 1 Target visibility comparison test

[0115]

[0116] Figure 5 A schematic diagram showing the tracking accuracy results of a slow, small target according to an embodiment of the present invention.

[0117] The error accuracy of UAV tracking low, slow and small targets is analyzed, such as Figure 5 As shown. The figure shows the trajectory tracking accuracy effect on different coordinate axes when the drone tracks a low, slow and small target, as well as the error changes of each coordinate axis. In the tracking of the X-axis and Y-axis, the drone follows the target trajectory more accurately, with a small error margin. There is a certain phase lag because the drone is required to maintain a safe distance of 2m from the low, slow and small target during design. The height change of the drone on the Z-axis was initially affected by the disturbance and showed obvious fluctuations, but it quickly stabilized after the initial fluctuations, and was finally able to maintain a good height near the target. Figure 5 The last figure shows the coordinate errors of the drone on the X, Y, and Z axes and the total error. It can be seen that in the initial stage, the Z-axis error of the drone is large, and then quickly converges to a smaller range. The errors of the X and Y axes show periodic fluctuations. This is because the set target moves in a circle, and the drone always maintains a safe distance of 2m from the low, slow, and small target, so the errors of the X and Y axes are always a sine curve with an amplitude of about 2m. During the overall tracking process, the error can gradually converge and remain within a small range of 2m, showing the robustness and control accuracy of the system. From the results, although there are certain errors in the early stage of flight due to disturbances and slow control response, the drone can eventually achieve accurate tracking of low, slow, and small targets through the coupling of visual perception information and tracking control, proving the effectiveness and robustness of the control system in dynamic and complex environments.

[0118] The present invention realizes the deep integration of visual perception information and flight control by combining deep learning target detection algorithm with model predictive control, solves the problems of response delay and insufficient stability caused by the separation of perception and control in traditional methods, and effectively improves the stability and anti-interference ability of multi-UAV collaborative tracking tasks based on the real-time disturbance compensation strategy of Gaussian process regression.

[0119] Example 2

[0120] The present disclosure provides an electronic device, which includes: a memory storing executable instructions; a processor running the executable instructions in the memory to implement the above-mentioned integrated perception and control method for drones to intercept low, slow and small targets.

[0121] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0122] The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0123] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory.

[0124] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0125] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0126] Example 3

[0127] An embodiment of the present disclosure provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the integrated perception and control method for drones to intercept low, slow and small targets is implemented.

[0128] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of each embodiment of the present disclosure are executed.

[0129] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0130] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any given examples.

[0131] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A perception and control integrated method for UAV to intercept low, slow and small targets, characterized in that: include: Build a control solution based on the drone perception-control integrated interception system; Build the target detection network BiGNet; Design a tracking and interception controller for low, slow and small targets.

2. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 1, wherein: The control scheme includes: Obtain original image information and identify the target drone through an aerial target detector; The perception information is integrated into the tracking controller for low, slow and small targets, and combined with the tracking disturbance compensator, a model predictive controller is constructed to control the UAV.

3. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 1, wherein: Building the target detection network BiGNet includes: A bidirectional feature pyramid network is used as the feature fusion mechanism to determine the feature fusion process; For the fusion process of the middle layer features, the features of each layer are combined through a weighted mechanism to determine the fusion features; The GhostConv module is introduced in BiGNet, which uses half of the features for convolution operations and concatenates the generated features with the original features to create redundant features.

4. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 3, wherein: The feature fusion process is: Among them, w i ≥0 is the learned weight, I i is the input feature map, and ε is a stable numerical coefficient.

5. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 3, wherein: The fusion features are: P i Td is the intermediate feature of the i-th layer, P i Out is the output feature of the i-th layer, Conv is a depth-wise separable convolution operation, and Resize is an upsampling or downsampling operation.

6. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 1, wherein: The design of the tracking and interception controller for low, slow and small targets includes: Visual perception information is introduced into the MPC optimization framework, and the cost functions of target tracking and disturbance compensation are added for joint optimization to obtain a tracking controller for low, slow and small targets.

7. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 6, wherein: The tracking controller for low, slow and small targets is: in, 8. The integrated perception and control method for drones to intercept low, slow and small targets according to claim 1, wherein: Also includes: By introducing disturbance conditions, the anti-disturbance ability and tracking stability of the tracking controller for low, slow and small targets in complex environments are verified.

9. An electronic device, characterized in that: The electronic device comprises: A memory storing executable instructions; A processor, wherein the processor runs the executable instructions in the memory to implement the integrated perception and control method for unmanned aerial vehicle intercepting low, slow and small targets as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the perception and control integration method for unmanned aerial vehicle intercepting low, slow and small targets as described in any one of claims 1-8.

Images