Anti-unmanned aerial vehicle system and method for training anti-unmanned aerial vehicle system
By using the decision-making model of storage anti-UAV agents and reinforcement learning algorithms in the anti-UAV system, the agents are trained for photoelectric detection, solving the problem of traditional methods identifying and tracking drones in complex environments, and achieving intelligence The risk of false alarm and missed alarm detection is reduced.
Patent Information
- Application Number
- CN202410932078.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
AI Technical Summary
Existing anti-UAV systems are difficult to achieve accurate identification, accurate tracking and flexible interference with drones in complex and changing battlefield environments. Traditional radar and radio-optical detection have problems such as high cost, poor portability and false alarm leakage. .
The decision-making model of the stored anti-UAV agent is adopted, including the feedback information generation module, the control information generation module and the designated direction search module. Through photoelectric detection results evaluation and confidence commands, the gimbal camera is controlled for automatic search and capture without external guidance. Signals are used to train the agents to achieve intelligent and unattended early warning detection using reinforcement learning algorithms.
It realizes photoelectric capture under radar or radio-guided signal conditions, or photoelectric search under no external guidance information, reducing false alarms and missing alarms, and improving the accuracy and automation level of drone early warning detection.
Smart Images

Figure CN118936241A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an anti-UAV system and a method for training the anti-UAV system. Background Art
[0002] With the rapid development of drone technology, its application in civil and military fields is becoming more and more extensive. Drone systems have become a new threat to battlefield air defense with their excellent reconnaissance and strike capabilities. In order to deal with the threat of drones, anti-drone technology has become an important technical field to ensure national and social security.
[0003] The application of anti-UAV technology can effectively detect, track, interfere with and destroy enemy UAVs, thereby reducing the enemy's reconnaissance and attack capabilities and improving our security capabilities. Anti-UAV technology is mainly divided into two parts: early warning and disposal. Among them, early warning mainly completes the detection and identification of UAVs, which is the most critical part of the anti-UAV system.
[0004] At present, anti-UAV systems mainly use efficient detection equipment and technical means to monitor and accurately identify UAVs in real time, providing the necessary target location, motion status, identity and other information for subsequent interference. However, current anti-UAV systems mostly use traditional radar, radio and photoelectric detection to achieve early warning. These methods have certain limitations and cannot adapt to complex and changing battlefields and application environments.
[0005] For example, radar detection, as an active detection method, can provide high detection accuracy, but due to its high cost and difficulty in portability, it is subject to certain limitations in practical applications; in addition, radar is prone to false alarms and may miss alarms for low-speed and stationary targets, further affecting the stability of its performance. Radio plus photoelectric detection is affected by the surrounding electromagnetic environment, and there is still much room for improvement in the indication accuracy and recognition accuracy of low, small and slow targets.
[0006] Therefore, how to accurately identify, accurately track and flexibly interfere with drones to achieve intelligent, unmanned drone early warning detection is the technical problem to be solved by the present invention. Summary of the invention
[0007] The purpose of the present invention is to provide an anti-UAV system and a method for training an anti-UAV system, which can not only accurately identify, accurately track and flexibly interfere with UAVs, but also realize intelligent and unmanned UAV early warning detection.
[0008] According to one aspect of the present invention, at least one embodiment provides an anti-UAV system, comprising: a memory, suitable for storing an anti-UAV intelligent agent decision model, the anti-UAV intelligent agent decision model comprising a feedback information generation module, a control information generation module and a specified position search module, the feedback information generation module being suitable for storing a plurality of instructions for evaluating guidance information based on photoelectric detection results and giving the confidence of the guidance information, the control information generation module being suitable for storing a plurality of instructions for generating an optimal search position based on the guidance information and the confidence, the specified position search module being suitable for determining the control amount of a gimbal camera according to the optimal search position; and one or more processors being suitable for loading and executing various instructions of the memory.
[0009] According to another aspect of the present invention, at least one embodiment also provides a method for training an anti-UAV system, which is applied to the above-mentioned anti-UAV system, wherein the one or more processors construct a feedback information network based on multiple instructions stored in a feedback information generation module and a control information network based on multiple instructions stored in a control information generation module, including: obtaining first guidance information of a current frame; sending the first guidance information and the second guidance information of a historical frame into the feedback information network to obtain a first confidence level of the current frame, sending the first confidence level and the first guidance information into the control information network to obtain a first optimal search position of the current frame, and determining a first control amount of a gimbal camera according to the first optimal search position; evaluating the first control amount, and optimizing the feedback information network and the control information network.
[0010] According to another aspect of the present invention, at least one embodiment also provides a method based on optoelectronic detection of an anti-UAV system, which is applied to the above-mentioned anti-UAV system, including: obtaining second guidance information provided by the anti-UAV system; controlling the optoelectronic device to rotate the field of view based on the second guidance information, and if there is a UAV in the field of view, capturing the UAV and performing optoelectronic tracking, or if there is no UAV in the field of view, searching for and capturing the UAV near the field of view and performing optoelectronic tracking.
[0011] According to another aspect of the present invention, at least one embodiment also provides an intelligent entity, including: a processor, suitable for implementing various instructions; and a memory, suitable for storing multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor: the method for training an anti-UAV system developed by the present invention and / or the method based on optoelectronic detection of an anti-UAV system developed by the present invention.
[0012] According to another aspect of the present invention, at least one embodiment further provides a computer-readable non-volatile storage medium storing computer program instructions, which, when the computer executes the program instructions, executes: the method for training an anti-UAV system developed by the present invention and / or the method based on optoelectronic detection of an anti-UAV system developed by the present invention.
[0013] The present invention trains the anti-UAV intelligent agent model in the above manner to realize the photoelectric capture of the anti-UAV early warning system under the conditions of radar or radio guidance signals, or the photoelectric search under the conditions of no external guidance information. By training the reinforcement learning algorithm model, the intelligent agent can gradually learn and master the effective action methods of searching and capturing targets, and realize intelligent and unattended UAV early warning detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0015] Figure 1 is a schematic diagram of an anti-UAV system according to an embodiment of the present invention;
[0016] Figure 2 is a schematic diagram of intelligent agent decision making in a multi-UAV scenario according to an embodiment of the present invention;
[0017] Figure 3 is a flow chart of a method for training an anti-UAV system according to an embodiment of the present invention;
[0018] Figure 4 is a schematic diagram of a preset algorithm A according to an embodiment of the present invention;
[0019] Figure 5 is a flow chart of a method for photoelectric detection based on an anti-UAV system according to an embodiment of the present invention;
[0020] Figure 6 is a two-dimensional schematic diagram of common guidance situations and controls according to an embodiment of the present invention;
[0021] Figure 7 is a two-dimensional schematic diagram of common guidance situations and controls according to an embodiment of the present invention;
[0022] Figure 8 It is a two-dimensional schematic diagram of common guidance situations and controls according to an embodiment of the present invention. DETAILED DESCRIPTION
[0023] The technical solution of the present invention will be described clearly and completely below 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 creative work are within the scope of protection of the present invention.
[0024] Generally speaking, photoelectric detection has high tracking accuracy, intuitive observation and identification, and strong interpretability, and can complement radar and radio detection very well. However, the field of view of photoelectric detection is small, and its ability to autonomously search and find targets is weak. In practical applications, skilled personnel are often required to perform manual control to achieve accurate detection effects, which increases the cost of use and the complexity of operation to a certain extent. Therefore, the inventor hopes to construct the anti-UAV system (or "UAV early warning system") of the present invention with the help of photoelectric detection technology.
[0025] At least one embodiment of the present invention provides an anti-UAV system (or "UAV early warning system"), such as Figure 1 As shown, the anti-UAV system may include: a memory 101, one or more processors 103, an optoelectronic device 105 equipped with a gimbal camera, and a guidance device 107.
[0026] The memory 101 is suitable for storing the anti-UAV intelligent agent decision model, which includes a feedback information generation module 1011, a control information generation module 1013 and a specified position search module 1015. The feedback information generation module 1011 is suitable for storing multiple instructions for evaluating guidance information based on photoelectric detection results and giving the confidence of the guidance information. The control information generation module 1013 is suitable for storing multiple instructions for generating the best search direction based on the guidance information and the confidence. The specified position search module 1015 is suitable for determining the control amount of the gimbal camera according to the best search direction.
[0027] The memory 101 can be various applicable storage devices, such as non-volatile storage devices, including but not limited to magnetic storage devices, semiconductor storage devices, optical storage devices, etc., and can be arranged as a single storage device, a storage device array or a distributed storage device, which is not limited by the embodiments of the present invention.
[0028] The anti-UAV agent decision model is also called the confidence-aware anti-UAV agent decision model: (1) The feedback information module 1011 is responsible for perception and decision-making, and receives the guidance information from the guidance device 107. and the photoelectric detection result V given by the photoelectric device 105 optical , evaluate the guidance device 107 and provide guidance information for this round The confidence level V confidence∈[0,1]. (2) The control information generation module 1013 and the designated direction search module 1015 work together to perform the action task. and the corresponding confidence V confidence Based on this, the control information generation module 1013 will output the next best search direction Then the designated position search module 1015 searches for the next best position according to the preset algorithm A and Calculate the control amount of the gimbal camera (OE) With the confidence V confidence The output strategy of the control information generation module 1013 needs to be adjusted accordingly: from completely relying on the instrument to detect the direction at the beginning, gradually changing to focusing only on the area near the direction, and finally exploring various possible directions. The advantage of this design is that algorithm A can adopt fixed rules without letting the agent learn, reducing the difficulty of training convergence.
[0029] The anti-UAV agent decision model can also provide good support for the detection of multiple UAV scenarios. For example, the anti-UAV agent decision model is simply adjusted to adapt to the multi-UAV scenario: when new guidance information appears, such as radar detecting new targets, adding new guidance equipment, etc., the present invention will instantiate a new feedback information generation module to be responsible for evaluating the confidence of the guidance information. Figure 2 First, assume that m feedback information generation modules have been instantiated. According to the number of existing optoelectronic devices n, the present invention will instantiate n control information generation modules Similarly, n designated position search modules are instantiated. Then, the present invention will select the guidance information based on the confidence level, that is, select the guidance information from the corresponding m confidence levels. Select the n instances with the largest output confidence values in, Finally, the present invention will be confidence And the corresponding guidance information Provided to the corresponding control information module Control information generated by it The corresponding optoelectronic devices are controlled through the one-to-one designated position search modules. At this point, the detection task of multiple drone scenes is completed.
[0030] One or more processors 103 are suitable for loading and executing various instructions of the memory 101. The one or more processors 103 can be various applicable processors, for example, implemented in the form of a central processing unit, a microprocessor, an embedded processor, etc., and can adopt an architecture such as X86, ARM, etc.
[0031] The optoelectronic device 105 equipped with a gimbal camera is suitable for receiving the gimbal camera (OE) control quantity provided by the designated position search module 1013, and is suitable for providing the optoelectronic detection result of the gimbal camera to the feedback information generation module 1011.
[0032] The guidance device 107 is adapted to provide guidance information to the feedback information generation module 1011 and the control information generation module 1013, wherein the guidance device 107 includes a radar and / or a radio detection device.
[0033] It will be understood by those skilled in the art that the structure of the anti-UAV system described above is only for illustration and does not limit the structure of the system. Figure 1 More or fewer components (such as transmission devices) shown in . The above-mentioned transmission device is used to receive or send data via a network. The network includes a wired network and a wireless network. The wireless network includes, but is not limited to, a wide area network, a metropolitan area network, a local area network or a mobile data network. Typically, the mobile data network includes, but is not limited to, a global mobile communication (GSM) network, a code division multiple access (CDMA) network, a wideband code division multiple access (WCDMA) network, a long-term evolution (LTE) communication network, a WIFI network, a ZigBee network, a network based on Bluetooth technology, etc. Different types of communication networks may be operated by different operators. The type of communication network does not constitute a limitation on the embodiments of the present invention. In one example, the transmission device is a radio frequency (RF) module, which is used to establish communication between the guidance device and the anti-UAV intelligent agent decision model by wireless means.
[0034] Through the above implementation, the present invention stores an anti-UAV agent decision model (or referred to as an "anti-UAV agent decision model" or "anti-UAV confidence perception agent decision module based on reinforcement learning") in the memory of the anti-UAV system, and realizes the photoelectric capture of the anti-UAV early warning system under the conditions of radar or radio guidance signals, or the photoelectric search under the conditions of no external guidance information by training the anti-UAV agent model. The present invention can enable the anti-UAV early warning system (or referred to as "agent") to gradually learn and master effective methods of searching and capturing targets by training the anti-UAV agent decision model, and realize intelligent, unmanned UAV early warning detection. This innovative anti-UAV agent decision model framework can provide a new solution for the development of anti-UAV early warning technology.
[0035] In the above operating environment, at least one embodiment of the present invention provides a method for training an anti-UAV system, which is applied to the above anti-UAV system: one or more processors 103 construct a feedback information network based on multiple instructions stored in the feedback information generation module 1011, and construct a control information network based on multiple instructions stored in the control information generation module 1013. Figure 3 As shown, a flowchart of a method for training an anti-UAV system is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical sequence is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here. The method may include the following steps:
[0036] Step S301, obtaining first guidance information of the current frame;
[0037] Step S303, sending the first guidance information and the second guidance information of the historical frame into the feedback information network to obtain the first confidence of the current frame, sending the first confidence and the first guidance information into the control information network to obtain the first best search orientation of the current frame, and determining the first control amount of the gimbal camera according to the first best search orientation;
[0038] Step S305, evaluating the first control variable, and optimizing the feedback information network and the control information network.
[0039] The present invention trains the anti-UAV intelligent agent model in the above manner to realize the photoelectric capture of the anti-UAV early warning system under the conditions of radar or radio guidance signals, or the photoelectric search under the conditions of no external guidance information. By training the reinforcement learning algorithm model, the intelligent agent can gradually learn and master the effective action methods of searching and capturing targets, and realize intelligent and unattended UAV early warning detection.
[0040] In step S301, the first guidance information of the current frame is obtained. For example, the trajectory of the drone in the current frame is collected. in, The horizontal field of view angle represents the real position of the drone in the i-th frame, Indicates the pitch angle of the real position of the drone in the i-th frame, Represents the distance of the real position of the UAV in the i-th frame; Get the first guide information of the current frame Among them, θ k1 Indicates the allowable deviation angle threshold of the horizontal field of view set by the guidance device. Indicates the allowable deviation angle threshold of the elevation field of view set by the guidance device. represents the horizontal field of view angle guided by the guiding device in the i-th frame, Indicates the pitch field angle guided by the guiding device in the i-th frame.
[0041] Optionally, based on the UAV flight trajectory collected in advance, the UAV 3D coordinate position of the current frame (such as the i-th frame) is converted into spherical coordinates relative to the camera coordinate system. And according to Generates guidance information given by the guidance device that simulates the current frame
[0042] In addition, in the process of studying the feedback information generation and control information generation logic, the inventors realized that it was significantly difficult to fully present this process in the form of specific code, so they chose to use neural networks to fit these two modules (the advantage of neural networks is that they can autonomously extract key features and establish a mapping between input and output through continuous training); in order to train the neural network, the inventors took the target truth value Adding Noise The method is used to simulate the boot device to give the boot information. The formula is: The advantage of this method is that by adjusting the noise The maximum value is used to directly simulate different detection situations of the optoelectronic device 105 (see the corresponding embodiment of the method based on optoelectronic detection of the anti-UAV system).
[0043] In step S303, the first guidance information and the second guidance information of the historical frame are sent to the feedback information network to obtain the first confidence of the current frame, the first confidence and the first guidance information are sent to the control information network to obtain the first optimal search direction of the current frame, and the first control amount of the gimbal camera is determined according to the first optimal search direction.
[0044] For example, the detection information of the current frame and the historical frame, that is, the "state" observed in reinforcement learning, is sent to the feedback information network to obtain the confidence of the current frame pass Get the current frame Among them, Net r represents the feedback information network, represents the hyperparameter of the horizontal field of view guided by the device in the previous k frames, represents the hyperparameter of the elevation field of view guided by the device in the previous k frames, The hyperparameter representing the photoelectric detection results of the photoelectric device in the previous k frames, represents the photoelectric detection result of the photoelectric device in the i-th frame, represents the first confidence of the i-th frame.
[0045] For example, the confidence of the current frame and guide information Send to the control information network to get the direction to be searched in the current frame pass get Among them, Net g Represents the control information network, Indicates the optimal search horizontal field of view of the optoelectronic device in the i-th frame, Indicates the optimal search pitch field angle of the optoelectronic device in the i-th frame.
[0046] For example, using the location to be searched and the current camera focal length f cur and direction Use the preset algorithm A (such as Figure 4 As shown in the figure, we can calculate the change of the gimbal position Δθ at this moment. And the focal length change Δf, which is the "action" performed in reinforcement learning. (1) Calculate the horizontal and elevation field of view angles of the current optoelectronic device w, h are the width and height of the area array imaging sensor, respectively, and f is the focal length of the optoelectronic device. (2) Field of view is θ range =[θ cur -θ fov ,θ cur +θ fov ], (3) If the direction to be searched is within the field of view, Then increase the focal length, output action Δθ=0, Δf=1; otherwise, it means the target is not in the field of view, and you need to turn towards the target and reduce the focal length. and / or Output action Among them, θ range Indicates the horizontal field of view of the optoelectronic device at this time. Indicates the pitch angle range of the optoelectronic device at this time, θ cur Indicates the horizontal field of view of the optoelectronic device at this time, represents the pitch angle of the optoelectronic device at this time, and the first control quantity includes Δθ, and Δf, Δθ represent the change in the horizontal field of view of the optoelectronic device. It represents the change of the pitch field angle of the optoelectronic device, and Δf represents the change of the focal length of the optoelectronic device.
[0047] In step S305, the first control amount is evaluated, and the feedback information network and the control information network are optimized. For example, the first control amount is evaluated according to the gimbal information and the coordinates of the drone in the current frame; the parameters of the feedback information network and the control information network are optimized using a model-free reinforcement learning algorithm, wherein the model-free reinforcement learning algorithm includes a PPO algorithm and / or a SAC algorithm; and multiple instructions of the feedback information generation module and the control information generation module are updated based on the parameters.
[0048] Optionally, the action is evaluated based on the gimbal information and drone coordinates of the current frame, i.e., the "reward" required for optimization in reinforcement learning. Record "state", "action", and "reward". If the number of records has reached the threshold for optimizing the model once, use the existing typical model-free reinforcement learning algorithm (such as PPO, SAC algorithm) to optimize the parameter part of the agent, i.e., the feedback information network and the control information network.
[0049] The present invention uses the above method to realize the automation of drone search tasks. First, based on the actual application scenario, the task is reasonably modeled; then, in order to ensure the application effect of the agent in different scenarios, a confidence-aware anti-drone agent decision model is designed; finally, the training of the agent is completed through reinforcement learning. It can be better applied to unmanned control pan-tilt camera (OE) to capture drones (UAV) under noisy guidance information.
[0050] In the above operating environment, at least one embodiment of the present invention provides a method based on photoelectric detection of an anti-UAV system, which is applied to the above anti-UAV system, such as Figure 5 As shown, including:
[0051] Step S501, obtaining second guidance information provided by the anti-UAV system;
[0052] Step S503, based on the second guidance information, the optoelectronic device is controlled to rotate the field of view, if there is a drone in the field of view, the drone is captured and optoelectronically tracked, or if there is no drone in the field of view, the drone is searched and captured near the field of view and optoelectronically tracked.
[0053] It should be noted that the working process of the anti-UAV system of the present invention is as follows: (1) the guiding device detects the UAV (also referred to as the "target") and provides guiding information; (2) the optoelectronic device rotates the field of view under the guidance of the guiding information to check whether there is an UAV target in the area indicated by the guiding information; (3) if there is a target in the field of view of the optoelectronic device, the target is captured through image detection and tracking algorithms, and optoelectronic tracking is performed; (4) if the target is not found in the field of view of the optoelectronic device, a search is carried out near the field of view range, and step (3) is entered; (5) if the optoelectronic device searches within a certain range and no target is found in the field of view, step (2) is entered. (6) until the target is found or the current search task is abandoned.
[0054] For example, let the target true position be The direction given by the guidance device is The target angular position output by the optoelectronic device is The focal length and adjustment range of the visible light or infrared camera in the optoelectronic device are f∈[f min ,f max ]. According to the pinhole imaging principle, the relationship between the focal length and the field of view of the optical system is: When the focal length is significantly different from the size of the imaging array, the field of view can be approximated by the inequality in the above formula, that is, the field of view angle is inversely proportional to the focal length, and the larger the focal length, the smaller the field of view. Therefore, the adjustment range of the horizontal and elevation field of view angles corresponding to the optoelectronic device is θ∈[θ min ,θ max ],
[0055] When the distance of the target meets the prerequisite for the photoelectric device to detect the target, that is, the distance between the target and the photoelectric device R>R optical Johnson's criterion finds pixel size based on the target's true position Leading angle position and the field of view of optoelectronic devices The relationship between the three, the optoelectronic equipment detects targets in a specific area under the guidance of guidance information, which can be divided into the following three situations, such as Figure 6 , 7 As shown in Figure 8, the five-pointed star represents the true position of the target, and the origin represents the instrument detection position. Figure 6 When the guidance device can accurately locate the UAV, the target's true angular position With guide angle position Within the range of the field of view of the photoelectric detection system focal length adjustment, the formula is expressed as |θ GT -θ Guide |≤θ min , Figure 7When there is a certain error in the target positioning accuracy of the guidance device, the guidance angle position The actual angular position of the drone A threshold range near The formula is θ min <|θ GT -θ Guide |≤θ k1 , Figure 8 When the target positioning accuracy of the guidance device is poor, the guidance angle position The actual angular position of the drone There is a serious deviation, exceeding the set allowable deviation angle threshold range The formula is expressed as θ k1 <|θ GT -θ Guide |, In an ideal state, the present invention will automatically or manually control the pan / tilt of the optoelectronic device to turn directly to the direction indicated by the guidance information, and gradually adjust the focal length (increase or decrease the focal length). Once the focal length is adjusted to a suitable range, the drone target can be observed in the image.
[0056] Although this simple rule has extremely high requirements on the performance of the environment and optoelectronic equipment, the present invention can gradually learn and master the effective action methods of searching and capturing targets by training the anti-UAV intelligent agent decision model, which greatly reduces the occurrence of false alarms and missed alarms, and realizes intelligent and unattended UAV early warning detection.
[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. An anti-UAV system, characterized in that: include: A memory, adapted to store an anti-UAV intelligent agent decision model, the anti-UAV intelligent agent decision model comprising a feedback information generation module, a control information generation module and a specified position search module, the feedback information generation module being adapted to store a plurality of instructions for evaluating guidance information based on photoelectric detection results and giving a confidence level of the guidance information, the control information generation module being adapted to store a plurality of instructions for generating an optimal search position based on the guidance information and the confidence level, and the specified position search module being adapted to determine a gimbal camera control amount according to the optimal search position; One or more processors are adapted to load and execute various instructions from the memory.
2. The anti-UAV system according to claim 1, characterized in that: Also includes: The optoelectronic device equipped with a gimbal camera is suitable for receiving the gimbal camera control amount provided by the specified position search module, and is suitable for providing the optoelectronic detection result of the gimbal camera to the feedback information generation module.
3. The anti-UAV system according to claim 1, characterized in that: Also includes: A guidance device is adapted to provide guidance information to the feedback information generation module and the control information generation module, wherein the guidance device includes a radar and / or a radio detection device.
4. A method for training an anti-UAV system, applied to the anti-UAV system according to any one of claims 1 to 3, characterized in that: The one or more processors construct a feedback information network based on a plurality of instructions stored in the feedback information generation module and construct a control information network based on a plurality of instructions stored in the control information generation module, including: Get the first guide information of the current frame; Sending the first guidance information and the second guidance information of the historical frame into a feedback information network to obtain a first confidence of the current frame, sending the first confidence and the first guidance information into a control information network to obtain a first optimal search orientation of the current frame, and determining a first control amount of the gimbal camera according to the first optimal search orientation; The first control quantity is evaluated, and the feedback information network and the control information network are optimized.
5. The method for training an anti-UAV system according to claim 4, wherein the first guidance information is It is characterized in that Obtaining the first guide information of the current frame includes: Collect the drone trajectory of the current frame in, The horizontal field of view angle represents the real position of the drone in the i-th frame, Indicates the pitch angle of the real position of the drone in the i-th frame, The distance representing the real position of the UAV in the i-th frame; according to Get the first guide information of the current frame Among them, θ k1 Indicates the allowable deviation angle threshold of the horizontal field of view set by the guidance device. Indicates the allowable deviation angle threshold of the elevation field of view set by the guidance device. represents the horizontal field of view angle guided by the guiding device in the i-th frame, Indicates the elevation field of view guided by the guiding device in the i-th frame.
6. The method for training an anti-UAV system according to claim 5, wherein the first confidence level is It is characterized in that Sending the first guide information and the second guide information of the historical frame into the feedback information network to obtain the first confidence of the current frame includes: pass Get the current frame Among them, Net r represents the feedback information network, represents the hyperparameter of the horizontal field of view guided by the device in the previous k frames, represents the hyperparameter of the elevation field of view guided by the device in the previous k frames, The hyperparameter representing the photoelectric detection results of the photoelectric device in the previous k frames, Represents the photoelectric detection result of the photoelectric device in the i-th frame.
7. The method for training an anti-UAV system according to claim 6, wherein the first optimal search position is It is characterized in that Sending the first confidence level and the first guidance information into a control information network to obtain a first optimal search position of the current frame includes: pass get Among them, Net g Represents the control information network, Indicates the optimal search horizontal field of view of the optoelectronic device in the i-th frame, Indicates the optimal search pitch field angle of the optoelectronic device in the i-th frame.
8. The method for training an anti-UAV system according to claim 7, wherein the first control variable comprises Δθ, and Δf, Δθ represent the change in the horizontal field of view of the optoelectronic device. represents the change in the pitch angle of the optoelectronic device, Δf represents the change in the focal length of the optoelectronic device, and is characterized in that: Determining a first control amount of the gimbal camera according to the first optimal search position includes: like Then output or like and / or Then output Among them, θ range Indicates the horizontal field of view of the optoelectronic device at this time. Indicates the pitch angle range of the optoelectronic device at this time, θ cur Indicates the horizontal field of view of the optoelectronic device at this time, Indicates the pitch field angle of the optoelectronic device at this time.
9. The method for training an anti-UAV system according to claim 4, characterized in that: Evaluating the first control quantity and optimizing the feedback information network and the control information network includes: Evaluate the first control amount according to the gimbal information and the coordinates of the drone in the current frame; Optimizing parameters of the feedback information network and the control information network using a model-free reinforcement learning algorithm, wherein the model-free reinforcement learning algorithm includes a PPO algorithm and / or a SAC algorithm; A plurality of instructions of the feedback information generation module and the control information generation module are updated based on the parameters.
10. A method based on photoelectric detection of an anti-UAV system, applied to an anti-UAV system as claimed in any one of claims 1 to 3, characterized in that: include: Obtaining the second guidance information provided by the anti-UAV system; Based on the second guidance information, the optoelectronic device is controlled to rotate the field of view. If there is a drone in the field of view, the drone is captured and optoelectronically tracked. If there is no drone in the field of view, the drone is searched and captured near the field of view and optoelectronically tracked.
Citation Information
Patent Citations
Federated Kalman filtering-based method for fusing multilateration data and radar data
CN101655561A
Method for implementing complementation and fusion of image tracking measurement data and radar measurement data
CN104064057A
A distributed cooperative search method for multi-UAV based on information fusion
CN109299210A
Anti-unmanned aerial vehicle detection tracking interference system and photoelectric tracking system working method
CN110398720A
Unmanned control decision-making method for photoelectric system of unmanned ship
CN115755926A
