An unmanned aerial vehicle countermeasure system and method based on bionic intelligent capture

Abstract

The application is an unmanned aerial vehicle countermeasure system and method based on bionic intelligent capture, belonging to the technical field of unmanned aerial vehicle countermeasures, comprising a data acquisition module for acquiring multi-dimensional feature data of a target unmanned aerial vehicle and generating a comprehensive signal matrix, the application can effectively complement the respective detection advantages by weighting and fusing the data acquired by the three types of sensors, significantly reducing the influence of adverse weather, lighting conditions or electromagnetic interference on target detection; ensuring that the fused data can most accurately reflect the real position of the target, thereby providing high-quality input for the subsequent module; enabling the system to improve from knowing where the target is to understanding what the target is and its possible attitude, providing a key basis for subsequent accurate decision-making; ensuring that the model has high-precision state calculation and feature recognition capabilities.

Description

Technical Field

[0001] The present invention relates to the technical field of drone countermeasures, and in particular to a drone countermeasure system and method based on bionic intelligent capture. Background Art

[0002] Existing drone countermeasures face a range of challenges when dealing with small, high-speed, and highly maneuverable non-cooperative drones. Traditional countermeasures have limitations in perception, decision-making, and execution, resulting in low countermeasure efficiency and success rates.

[0003] Specifically, these limitations are reflected in:

[0004] Detection systems that rely on a single sensor type, such as purely optical or radar systems, are susceptible to interference from inclement weather, varying lighting conditions, or complex electromagnetic environments. This makes it difficult to provide stable, accurate, and all-weather perception of target drones, resulting in incomplete information acquisition. Traditional tracking algorithms typically focus solely on kinematic information such as the target's position and velocity, lacking effective identification of intrinsic properties such as target type, size, and posture. This prevents countermeasure systems from developing tailored interception strategies based on target characteristics. Most countermeasure systems respond based on the target's current or historical position, employing a pursuit-based interception strategy. This passive pursuit approach is ineffective against intelligent drones capable of evasive maneuvers, lacking effective prediction of the target's future trajectory and preventing efficient pre-positioned interception. Interception methods and parameters are often fixed, such as a single interceptor flight mode or a fixed launch timing and configuration for a capture net. This one-size-fits-all strategy is unsuitable for diverse interception scenarios and targets with varying characteristics, resulting in low interception success rates or poor aerodynamic and energy efficiency in specific scenarios.

[0005] The root cause of the above problems lies in the fragmentation of various technical links and the lack of data processing and decision-making capabilities, that is, the failure to integrate precise perception, deep understanding, forward-looking prediction and adaptive decision-making, making it difficult to cope with the growing dynamic and non-cooperative drone threats.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0007] The purpose of the present invention is to provide a drone countermeasure system and method based on bionic intelligent capture to solve the problems raised in the above background technology.

[0008] The technical solution of the present invention is as follows: it comprises a data acquisition module for acquiring multi-dimensional feature data of a target UAV and generating a comprehensive signal matrix;

[0009] a state estimation module, configured to receive the integrated signal matrix, calculate a motion state vector including target position and velocity, and extract an intrinsic feature vector including target type and posture;

[0010] a trajectory prediction module, configured to generate a predicted future position of the target UAV based on the motion state vector and the intrinsic feature vector of the historical sequence;

[0011] The countermeasure decision module is used to generate a flight control strategy and a capture net control strategy based on the future predicted position, the intercepting drone's own flight status and the intrinsic characteristic vector. The flight control strategy determines the flight mode and wingspan parameters for trajectory tracking, and the capture net control strategy determines the release instructions and net tension parameters for triggering capture.

[0012] Preferably, the data acquisition module is further used to:

[0013] Collect visible light intensity signals, infrared intensity signals, and radar echo intensity signals obtained by visible light sensors, infrared sensors, and millimeter-wave radar sensors as multi-source perception data;

[0014] The integrated signal matrix is ​​generated by performing weighted fusion on the multi-source perception data, and the integrated signal matrix is ​​used to provide input for the state estimation module.

[0015] Preferably, the state estimation module is further used to:

[0016] Constructing a falcon-like visual neural network model, and using the integrated signal matrix as input to the model;

[0017] Through the convolution and fully connected layer processing of the model, the motion state vector representing the three-dimensional spatial position and speed of the target is calculated, and the intrinsic feature vector representing the intrinsic properties of the target is extracted. The motion state vector and the intrinsic feature vector are used in the trajectory prediction module.

[0018] Preferably, the trajectory prediction module is further used to:

[0019] Construct a long short-term memory network model;

[0020] Inputting a feature sequence composed of the motion state vector at a historical moment and the intrinsic feature vector into the long short-term memory network model;

[0021] The model generates the future predicted position of the target UAV by learning the dynamic behavior of the target in consecutive frames, and the future predicted position is used to provide a decision basis for the countermeasure decision module.

[0022] Preferably, when determining the flight control strategy, the countermeasure decision module is further configured to:

[0023] Obtain target distance, relative speed, altitude difference and relative angle as flight status parameters;

[0024] Constructing a mode switching decision function to calculate the fitness values ​​of six flight modes: cruising, tracking, diving, circling, hovering, and capturing based on the flight state parameters;

[0025] The mode with the highest fitness value is selected as the flight mode, which is used for determining the subsequent wingspan parameters.

[0026] Preferably, the countermeasure decision module is further configured to:

[0027] Obtaining the selected flight mode and the flight speed of the intercepting UAV itself;

[0028] Constructing a wingspan optimization function to dynamically fine-tune the basic wingspan corresponding to the flight mode and the flight speed of the intercepting drone itself;

[0029] An optimal wingspan is calculated as the wingspan parameter, and the wingspan parameter is used to drive a dynamic wingspan adjustment mechanism to perform.

[0030] Preferably, when determining the flight control strategy, the countermeasure decision module is further configured to:

[0031] Obtaining the future predicted position generated by the trajectory prediction module and obtaining the current flight status of the intercepting UAV;

[0032] Constructing a trajectory tracking control law to calculate the total thrust and torque vector required to achieve tracking of the predicted future position;

[0033] The total thrust and torque vector is decomposed into thrust instructions for each independent fan, and the thrust instructions are used to execute the flight control strategy.

[0034] Preferably, when determining the delivery instruction, the countermeasure decision module is further configured to:

[0035] Obtaining the predicted future position, and calculating a predicted interception accuracy factor, a relative speed factor, a target maneuverability factor, and a target attitude factor to form a capture evaluation index set;

[0036] Constructing a capture probability evaluation model to calculate the probability of successful capture based on the capture evaluation indicator set;

[0037] The successful capture probability is compared with a preset confidence threshold, and when the successful capture probability exceeds the confidence threshold, the delivery instruction is generated.

[0038] Preferably, when determining the mesh tension parameter, the countermeasure decision module is further configured to:

[0039] parsing equivalent size information of the target from the intrinsic feature vector generated by the state estimation module;

[0040] An optimal tension calculation function is constructed to calculate the optimal mesh tension required to ensure reliable wrapping according to equivalent size information of the target;

[0041] The optimal grid tension is set as the mesh tension parameter, and the mesh tension parameter is used to control the stiffness of the magnetorheological material mesh.

[0042] The countermeasure decision module is further configured to:

[0043] Obtaining the future predicted position and the mesh tension parameter;

[0044] Constructing a mesh deployment trajectory optimization model, and inputting the mesh tension parameter as a physical constraint affecting the deployment behavior;

[0045] The initial speed and duration of the deployment mechanism are optimized to ensure that the center of the deployed mesh can accurately intercept the future predicted position.

[0046] A drone countermeasure method based on bionic intelligent capture includes the following steps:

[0047] Step 1: Multi-dimensional data perception: The multi-source sensor array of the data acquisition module collects the visible light, infrared and radar echo signals of the target UAV and performs weighted fusion to generate a comprehensive signal matrix that represents the multi-dimensional characteristics of the target;

[0048] Step 2: State calculation and intention recognition: The integrated signal matrix is ​​processed using the falcon-like visual neural network model built into the state estimation module to calculate the current three-dimensional spatial position, speed and other motion states of the target UAV, and simultaneously extract its type, posture and other intrinsic features;

[0049] Step 3: Future trajectory prediction: The motion state and intrinsic features of the historical time series are input into the long short-term memory network model of the trajectory prediction module to learn the dynamic behavior pattern of the target and generate its predicted position at the future moment;

[0050] Step 4: Countermeasure strategy generation. The countermeasure decision module makes a multimodal decision based on the target's predicted future position and intrinsic characteristics, combined with the intercepting drone's own flight status:

[0051] a) Determine the flight control strategy: Calculate the optimal flight modes and corresponding wingspan parameters, including cruise, tracking, and hovering, and generate the thrust and torque commands required for trajectory tracking;

[0052] b) Determine the capture net control strategy: evaluate the probability of successful capture and calculate the optimal instructions to trigger the capture net release and the net tension parameters required to ensure reliable capture;

[0053] Step 5: Integrated collaborative execution: Based on the generated strategy, the intercepting UAV is precisely controlled to perform trajectory tracking and maneuvering. When the probability of successful capture meets the preset threshold, the release and stiffness adjustment of the capture net are triggered, and the target UAV is finally captured.

[0054] The present invention provides a drone countermeasure system and method based on bionic intelligent capture through improvements, which has the following improvements and advantages compared with the existing technology:

[0055] 1. By weightedly fusing the data obtained by these three types of sensors, they can effectively complement their respective detection advantages and significantly reduce the impact of adverse weather, lighting conditions, or electromagnetic interference on target detection.

[0056] 2. Ensure that the fused data can most accurately reflect the true location of the target, thereby providing high-quality input for subsequent modules;

[0057] 3. This enables the system to move from knowing where the target is to understanding what the target is and its possible posture, providing a key basis for subsequent accurate decision-making;

[0058] 4. Ensure that the model has both high-precision state solution and feature recognition capabilities;

[0059] 5. It can learn the complex flight dynamics pattern of the target and predict its future spatial position based on historical information. This allows the intercepting drone to take a more optimal interception path and go directly to the target's predicted future position, realizing intelligent interception in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The present invention will be further explained below in conjunction with the accompanying drawings and Examples:

[0061] Figure 1 It is a flow chart of the system of the present invention. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0063] Example 1:

[0064] See also Figure 1,The present invention provides a technical solution of a UAV countermeasure system and method based on bionic intelligent capture, comprising: a data acquisition module for acquiring multi-dimensional feature data of a target UAV and generating a comprehensive signal matrix;

[0065] The state estimation module receives the integrated signal matrix, calculates the motion state vector containing the target position and velocity, and extracts the intrinsic feature vector containing the target type and posture;

[0066] The trajectory prediction module is used to generate the future predicted position of the target UAV based on the motion state vector and intrinsic feature vector of the historical sequence;

[0067] The countermeasure decision module is used to generate flight control strategy and capture net control strategy based on the future predicted position, the intercepted drone's own flight status and internal feature vector. The flight control strategy determines the flight mode and wingspan parameters for trajectory tracking, and the capture net control strategy determines the release instructions and net tension parameters used to trigger capture.

[0068] This embodiment provides a drone countermeasure system based on bionic intelligent capture. The system realizes effective detection, tracking, prediction and interception of target drones through the chain-like collaborative work of a data acquisition module, a state estimation module, a trajectory prediction module and a countermeasure decision module. The data acquisition module uses multi-source sensor fusion technology to provide accurate input for the state estimation module. The state estimation module draws on the biological vision mechanism to solve the target's real-time motion state and intrinsic properties for use by the trajectory prediction module. The trajectory prediction module uses a deep learning network to make a high-precision prediction of the target's future position. The countermeasure decision module comprehensively plans the optimal flight control strategy and capture network control strategy based on the predicted position and its own state. This design integrates perception, estimation, prediction and decision-making through the tight data flow and logic flow between modules, ensuring the efficiency and success rate of the countermeasure task against dynamic and non-cooperative drone targets.

[0069] Example 2

[0070] The data acquisition module is further used to:

[0071] Collect visible light intensity signals, infrared intensity signals, and radar echo intensity signals obtained by visible light sensors, infrared sensors, and millimeter-wave radar sensors as multi-source perception data;

[0072] By performing weighted fusion on multi-source perception data, a comprehensive signal matrix is ​​generated, which is used to provide input for the state estimation module.

[0073] The state estimation module is further used to:

[0074] Construct a falcon-like visual neural network model and use the integrated signal matrix as the input of the model;

[0075] Through the convolution and fully connected layer processing of the model, the motion state vector representing the three-dimensional spatial position and velocity of the target is calculated, and the intrinsic feature vector representing the intrinsic properties of the target is extracted. The motion state vector and intrinsic feature vector are used in the trajectory prediction module.

[0076] In this embodiment, the data acquisition module uses an integrated array of visible light sensors, infrared sensors, and millimeter-wave radar sensors to achieve all-weather, multi-dimensional information acquisition of target drones. This module collects raw signals from each sensor, including visible light intensity signals, infrared intensity signals, and radar echo intensity signals, to form multi-source perception data. The data acquisition module performs weighted fusion processing on the multi-source perception data to generate a comprehensive signal matrix. This fusion process adjusts the proportion of each sensor data in the final output using preset weight coefficients, enhancing the system's robustness in complex electromagnetic and meteorological environments.

[0077]

[0078] in, For the The integrated signal strength of each sensing unit, is the visible light intensity signal of the unit, is the infrared intensity signal, is the radar echo strength signal, is the pre-calibrated weight coefficient;

[0079] The pre-calibration process is as follows: In a controlled test environment with various weather and lighting conditions, such as sunny, cloudy, night, light rain, and light fog, a high-precision motion capture system, such as Vicon or OptiTrack, is used as a true reference to simultaneously record the precise 3D position of the target drone; at the same time, the system collects raw data from visible light, infrared, and millimeter wave radars. An optimization algorithm, such as the least squares method, is used to find the optimal set of weight coefficients. , whose goal is to minimize the error between the position calculated by the fused signal and the true position provided by the motion capture system; the objective function of the optimization process can be expressed as:

[0080]

[0081] in, and is the position and velocity calculated by the fusion signal solution, and and It is the true value provided by the high-precision motion capture system; and It is a preset weight coefficient used to balance the importance of position error and speed error in the optimization objective;

[0082] in, is the total number of samples collected, An abstract function that represents the complete process of calculating the motion state from the fused signal, including position and velocity. For the The true value position of the samples; this set of optimal weights It is solidified in the data acquisition module; : Represents the weight coefficients of visible light, infrared and millimeter wave radar sensor data in the fusion process; : The optimal weight coefficient obtained after calibration optimization; : The total number of test samples used to calibrate the weight coefficient; : An abstract function that represents the complete process of calculating the target position from the fused signal solution, which is used as a functional abstraction in the optimization objective function; :During the calibration process, the first The true location of the samples; :Respectively refers to Sampling time, by Visible light intensity signals, infrared intensity signals and radar echo intensity signals collected by each sensing unit;

[0083] The state estimation module receives the integrated signal matrix generated by the data acquisition module. This module has a pre-trained falcon-like visual neural network model built into it, which mimics the hierarchical processing mechanism of the raptor's visual cortex.

[0084] The model introduces a spatial attention module before the first convolution layer, which simulates the foveal mechanism in the falcon's visual system and can integrate the signal matrix The high information density areas in the image are given higher processing weights, so that the key areas containing the target subject can be processed first without significantly increasing the amount of calculation;

[0085] The training process is based on a large amount of historical drone flight data collected in laboratory and field environments, giving it powerful feature extraction and state resolution capabilities. The integrated signal matrix serves as the input to the falcon-like visual neural network model. After being processed by multiple convolutional and fully connected layers within the model, it simultaneously outputs two key pieces of information: a motion state vector and an intrinsic feature vector. The motion state vector accurately describes the target's current three-dimensional spatial position and velocity, while the intrinsic feature vector contains deep attribute information such as the target's type and posture. Together, these two constitute a complete description of the target's state and serve as the core input to the subsequent trajectory prediction module, laying the data foundation for accurately predicting the target's intention and trajectory.

[0086]

[0087] in, for The position component of the motion state vector at time , is the velocity component, is the intrinsic eigenvector at that moment, represents the falcon-like visual neural network model function, for The integrated signal matrix input at all times, and are the weight and bias parameters of the model respectively;

[0088] The falcon-like visual neural network model C is a multi-task convolutional neural network with the following detailed structure:

[0089] Input layer: receives the integrated signal matrix generated by the data acquisition module , whose dimensions are normalized to ;

[0090] Convolutional layer: contains 5 convolution blocks, each of which consists of a A convolutional layer with ReLU activation function, a batch normalization layer and a The maximum pooling layer is composed of

[0091] Fully connected layer: After the convolutional layer, two fully connected layers with 1024 neurons each are connected;

[0092] Output layer: Two branches are separated from the second fully connected layer:

[0093] Motion state branch: a linear output layer containing 6 neurons, used to regress and calculate the motion state vector composed of three-dimensional space position and velocity ;

[0094] Intrinsic feature branch: an output layer containing N neurons, where N is the total number of predefined object types and pose categories;

[0095] In this embodiment, , which includes 5 target type categories, such as small quadrotor, medium hexacopter, drone, fixed wing, unknown, and 10 attitude categories, such as level, hover, climb, dive, turn left, turn right, forward, backward, left flight, and right flight; the intrinsic feature branch of the model outputs a vector of length 15, corresponding to the probabilities of these 15 categories;

[0096] Use Softmax activation function to calculate intrinsic feature vector for classification ;

[0097] The model is trained on a dataset of more than 200,000 manually annotated drone flight images collected in various environments.

[0098] The dataset covers at least 15 common multi-rotor and fixed-wing drone models, captured in a variety of background and lighting conditions, including urban and rural areas, daytime, dusk, nighttime, and light rain and fog. Each image is annotated with not only the true 3D position and velocity of the target, but also 10 predefined attitude categories, such as level, climbing, and left turn, and five drone types, such as small quadrotors and medium hexacopters.

[0099] The training adopts Adam optimizer, and the learning rate is , the total loss function is the mean square error loss of the motion state branch Cross entropy loss with intrinsic feature branch The weighted sum of:

[0100]

[0101] The weight and It is set to 1.0 and 0.5 to balance the importance of regression and classification tasks; : refers to the entire falcon-like visual neural network model; :exist Time, input to the neural network model The integrated signal matrix; :exist At time t, the position vector output by the model; :exist At time , the velocity vector output by the model; :exist At this moment, the intrinsic feature vector output by the model contains information such as target type and posture; : The total loss function used when training the neural network model; : Mean squared error loss, used to evaluate the error of position and velocity regression tasks; : Cross entropy loss, used to evaluate the error of target type and pose classification tasks; : Used to balance the two loss functions and The weight coefficient of the proportion in the total loss.

[0102] Example 3

[0103] The trajectory prediction module is further used to:

[0104] Construct a long short-term memory network model;

[0105] Input the feature sequence composed of the motion state vector and the intrinsic feature vector at the historical moment into the long short-term memory network model;

[0106] The model generates the future predicted position of the target drone by learning the dynamic behavior of the target in consecutive frames. The future predicted position is used to provide a decision basis for the countermeasure decision module.

[0107] In this embodiment, the core of the trajectory prediction module is a constructed long-short-term memory network model. This model is specifically designed to process and predict time series data. Its internal gating mechanism can effectively learn long-term dependencies and is suitable for capturing the complex flight dynamics of drones. The trajectory prediction module combines the motion state vectors output by the state estimation module at multiple consecutive moments in the past with the intrinsic feature vectors to form a historical state-feature sequence. This sequence is provided as input to the long-short-term memory network model. By analyzing this sequence, the model learns the dynamic behavior patterns of the target in consecutive frames, such as its maneuvering intentions and flight inertia. After learning is completed, the model generates a predicted position of the target drone at a certain point in the future. This future predicted position is the key decision-making basis for the countermeasure decision-making module to formulate the interception strategy, and its accuracy directly affects the success or failure of the interception mission.

[0108]

[0109] in, For the future The target position prediction result of the step size, represents the long short-term memory network model function, To include the past The historical sequence of state vectors and feature vectors at each moment, For the current moment.

[0110] Specifically, the long short-term memory network model L is a two-layer stacked LSTM structure; the input feature sequence The motion state vector output by the state estimation module at the past m moments is With intrinsic eigenvector Each LSTM layer contains 256 hidden units. After the LSTM layer, a fully connected layer is connected, which outputs a vector of length 3, which is the target position for the next k steps. The model is also trained using the Adam optimizer, and the loss function is the mean square error between the predicted position and the true future position. : refers to the long short-term memory network model; : Contains the past The historical feature sequence of the state vector at each moment is used as the input of the LSTM model; : defines the input sequence the time step or history length; : defines the future time step that needs to be predicted; :The model predicts that the target is in the future Spatial location at a moment;

[0111] Example 4

[0112] The countermeasure decision module is further used to determine the flight control strategy:

[0113] Obtain target distance, relative speed, altitude difference and relative angle as flight status parameters;

[0114] Construct a mode switching decision function and calculate the fitness values ​​of six flight modes: cruise, tracking, dive, hover, hover, and capture based on flight state parameters;

[0115] The mode with the highest fitness value is selected as the flight mode, and the flight mode is used to determine the subsequent wingspan parameters.

[0116] The countermeasure decision module is further used to:

[0117] Get the selected flight mode and the intercepting drone's own flight speed;

[0118] Construct a wingspan optimization function, and dynamically fine-tune it based on the basic wingspan corresponding to the flight mode and the intercepting drone's own flight speed;

[0119] The optimal wingspan is calculated as the wingspan parameter, and the wingspan parameter is used to drive the wingspan dynamic adjustment mechanism to execute.

[0120] In this embodiment, the countermeasure decision module determines the flight control strategy by acquiring real-time flight state parameters between the interceptor and target aircraft, including target distance, relative speed, altitude difference, and relative angle. These parameters are input into a preset mode switching decision function. This function defines fitness evaluation rules for six preset flight modes: cruise, tracking, dive, hover, hover, and capture. Based on the input flight state parameters, the function calculates the fitness value of each mode in real time. The mode with the highest fitness value is selected as the current optimal flight mode.

[0121]

[0122] in, is the selected optimal flight mode, For the The fitness function of the modality, Represent target distance, relative speed, height difference and relative angle respectively;

[0123] After determining the flight mode, the countermeasure decision module further obtains the selected flight mode and the interceptor drone's own flight speed; this information is fed into the wingspan optimization function to calculate the optimal wingspan parameters. The core logic of this function is to make smooth dynamic fine-tuning based on the interceptor's own speed on the basic wingspan value corresponding to the selected mode; for example, in high-speed tracking mode, the wingspan will be appropriately reduced to reduce drag; in low-speed circling mode, the wingspan will be increased to provide more lift; the calculated optimal wingspan parameters are then sent to the wingspan dynamic adjustment mechanism to drive the physical wing surface to complete the deformation, so that the interceptor achieves the optimal aerodynamic performance in the current mission phase;

[0124]

[0125] in, is the calculated optimal wingspan, Current flight mode The corresponding basic wingspan, is the maximum adjustment amplitude under this mode, is the dimensionless sensitivity coefficient, is the interceptor's own flight speed, is the reference flight speed of this mode;

[0126] Fitness evaluation rules for some modes The definition is as follows:

[0127] Tracking Mode( ): Suitable for stable tracking scenarios with medium distance and low relative speed; the fitness function is designed as:

[0128]

[0129] As a preferred embodiment, a set of calibrated typical parameter values ​​are: weight , optimal tracking distance Meters, distance attenuation coefficient , relative velocity attenuation coefficient ;

[0130] in is the target distance, is the relative speed, is the optimal tracking distance, and are the preset weights and attenuation coefficients, ensuring that and Get high scores when

[0131] Diving mode ( ): It is suitable for scenarios where we have a significant height advantage and need to quickly approach the target. Its fitness function is designed as:

[0132]

[0133] As a preferred embodiment, a set of calibrated typical parameter values ​​are: weight , ideal initial height advantage Meters, height attenuation coefficient ;

[0134] in is the height difference, is the horizontal distance, This demonstrates the effectiveness of the dive angle. It is an ideal initial height advantage, To indicate the function, ensure that this mode is only considered when our side is above the target;

[0135] Hover Mode ( ): It is suitable for scenarios where the target is highly maneuverable and requires stable observation or waiting for the best attack opportunity. Its adaptability is determined by the acceleration or angular velocity of the target in a short period of time. Decide:

[0136]

[0137] As a preferred embodiment, a set of calibrated typical parameter values ​​are: weight , maneuverability sensitivity coefficient ;

[0138] That is, the more intense the target maneuvers, the higher the adaptability of the hovering wait. All weights and coefficients are calibrated through simulation and actual flight data; : No. Fitness evaluation function of the flight mode; : fitness functions of pursuit, dive and hover modes respectively; : Preset weight coefficient used in fitness function; : The preset attenuation or sensitivity coefficient used in the fitness function; : The current distance between the intercepting drone and the target drone; : preset optimal tracking distance; : The relative speed between the intercepting UAV and the target UAV; : The altitude difference between the intercepting UAV and the target UAV; : Preset ideal initial height advantage; : The current tangential acceleration of the target analyzed from historical data; : Indicator function, when the condition in the brackets is true, the function value is 1, otherwise it is 0;

[0139] Basic wingspan , Maximum adjustment range , adjust the sensitivity coefficient And the reference flight speed Is related to the selected flight mode Directly associated preset parameters; these parameters are stored in a lookup table. When a flight mode is selected, the controller retrieves the corresponding parameter group from the table and substitutes it into the wingspan optimization function for real-time calculation; : The optimal flight mode selected after fitness evaluation; : Optimal wingspan calculated in real time; : With flight mode The corresponding basic wingspan value; : The maximum adjustable range of the wingspan in the current mode; : The sensitivity coefficient of wingspan adjustment in the current mode; : The intercepted drone’s own flight speed; : The reference flight speed corresponding to the current flight mode;

[0140] Example 5

[0141] The countermeasure decision module is further used to determine the flight control strategy:

[0142] Get the future predicted position generated by the trajectory prediction module and get the current flight status of the intercepting drone;

[0143] Construct a trajectory tracking control law to calculate the total thrust and torque vector required to track the predicted future position;

[0144] The total thrust and torque vector is decomposed into thrust commands for each independent fan, and the thrust commands are used to execute the flight control strategy.

[0145] When determining the delivery instruction, the countermeasure decision module is further used to:

[0146] Obtain the future predicted position and calculate the predicted interception accuracy factor, relative speed factor, target maneuverability factor, and target attitude factor to form a capture evaluation index set;

[0147] Construct a capture probability evaluation model and calculate the probability of successful capture based on the capture evaluation indicator set;

[0148] Compare the probability of successful capture with a preset confidence threshold, and generate a release instruction when the probability of successful capture exceeds the confidence threshold;

[0149] The preset confidence threshold can be optimized by conducting a large number of capture experiments in a simulation environment and analyzing the relationship between the success rate and the false trigger rate. In a preferred embodiment, the threshold is set to 0.9;

[0150] In this embodiment, to perform flight control, the countermeasure decision module obtains the predicted future position generated by the trajectory prediction module and the interceptor's current flight state. These two pieces of information are input into a preset trajectory tracking control law. Based on optimal control theory, this control law calculates the total thrust and torque vector required to accurately drive the interceptor to the predicted target position. This total thrust and torque vector is decomposed into specific thrust commands for the four independent ducted fans through a pre-calibrated thrust allocation matrix. These commands ultimately drive each fan to adjust its speed, achieving precise control of the flight trajectory.

[0151]

[0152] in, is the required total thrust and torque vector, represents the trajectory tracking control law function, Predict the future position of the target, is the current state of the interceptor, is the current flight mode;

[0153] When determining the capture net release command, the countermeasure decision module also obtains the future predicted position and uses this to calculate the four key factors that constitute the capture evaluation index set: predicted interception accuracy factor, relative speed factor, target maneuverability factor, and target attitude factor. These factors quantify the geometric feasibility of interception, speed matching, target avoidance ability, and whether the attitude is conducive to capture. This index set is input into a pre-trained capture probability assessment model, which outputs a successful capture probability value. This probability value is compared with a preset confidence threshold. Only when the calculated successful capture probability exceeds the threshold will the system generate and issue the capture net release command. This decision-making mechanism based on quantitative evaluation avoids blind deployment and significantly improves capture efficiency and resource utilization.

[0154]

[0155] in, To successfully capture the probability, They are the four evaluation factors of predicted interception accuracy, relative speed, target maneuverability and target attitude. is the weight coefficient corresponding to each factor, which is obtained by studying a large number of successful and failed capture cases;

[0156] The four evaluation factors are quantified as follows:

[0157] Predicted interception precision factor ( ): Based on the predicted future position of the target The interception point we can reach The Euclidean distance between

[0158]

[0159] Relative velocity factor ( ): The relative speed is required to be as small as possible during capture. This factor is modeled using a Gaussian function. It reaches its peak value when it is zero;

[0160]

[0161] Target mobility factor ( ): Based on the target's current tangential acceleration extracted from the historical trajectory The more maneuverable the target, the more difficult it is to capture.

[0162]

[0163] Target attitude factor ( ): Based on the target intrinsic feature vector The attitude calculated in [1]. For example, when capturing a rotary-wing drone from directly above, its rotors are exposed to the greatest extent and its attitude is most favorable. This factor can be quantified as a lookup table value or calculated based on the angle between the intercept direction and the target coordinate system. For example, it can be defined as a function of the cosine of the angle between the intercept vector and the target's Z axis (vertically upward);

[0164]

[0165] The weight coefficient , in the total formula is The set of is obtained by fitting logistic regression on tens of thousands of simulation capture data;

[0166] : The final calculated probability of successful capture; : are the predicted intercept accuracy factor, relative speed factor, target maneuverability factor and target attitude factor respectively; : In the logistic regression model, all weight coefficients Collection : predicted future position of the target; : The interception point position that our interceptor can reach; : Predict the relative speed at the moment of interception; : tangential acceleration of the target; : intercept vector; : The Z axis of the target drone, the vertical upward unit vector, is used to determine the posture.

[0167] Example 6

[0168] When determining the network tension parameters, the countermeasure decision module is further used to:

[0169] Analyze the equivalent size information of the target from the intrinsic feature vector generated by the state estimation module;

[0170] The specific analysis process is as follows: The target type category with the highest confidence is identified; the system searches for the corresponding average equivalent size in a preset lookup table according to the type ,For example, if it is identified as a medium-sized six-rotor, its corresponding equivalent size is obtained from the lookup table as 0.8 meters;

[0171] Construct an optimal tension calculation function to calculate the optimal mesh tension required to ensure reliable wrapping based on the equivalent size information of the target;

[0172] The optimal mesh tension is set as the mesh tension parameter, which is used to control the stiffness of the magnetorheological material mesh.

[0173] The countermeasure decision module is further used to:

[0174] Obtain future predicted position and mesh tension parameters;

[0175] Construct a mesh deployment trajectory optimization model and input the mesh tension parameters as the physical constraints affecting the deployment behavior;

[0176] Optimize the initial speed and duration of the deployment mechanism to ensure that the center of the deployed mesh can accurately intercept the future predicted position.

[0177] In this embodiment, to determine the mesh tension parameter, the countermeasure decision module extracts the target's equivalent size information from the intrinsic eigenvector generated by the state estimation module. This information is input into an optimal tension calculation function. Based on the target's size, this function calculates the optimal mesh tension that ensures flexible wrapping to prevent the target from bouncing off while also providing sufficient toughness to absorb impact energy. The calculated optimal mesh tension is set as the mesh tension parameter. This parameter is used to control the capture net woven from magnetorheological material. By adjusting the magnetic field strength applied to the material, the mesh's stiffness is changed in real time, enabling adaptive capture of targets of varying sizes.

[0178]

[0179] in, is the calculated optimal mesh tension, represents the optimal tension calculation function, is the target equivalent size resolved from the intrinsic eigenvector;

[0180] After determining the tension, the countermeasure decision module further obtains the target's predicted future position and the determined net tension parameters. These two parameters are input into the net deployment trajectory optimization model. In this model, the net tension parameter serves as a key physical constraint because it directly affects the aerodynamic characteristics and morphological changes of the net during air deployment. The model uses an optimization algorithm to solve a set of optimal deployment mechanism parameters, mainly including the initial deployment speed and deployment duration. This set of parameters ensures that the center point of the deployed capture net accurately coincides with the target's predicted future position in time and space, while taking into account the influence of tension, thereby achieving precise interception.

[0181]

[0182] in, is the expansion parameter set to be optimized, initial speed, duration, expansion angle, is the predicted interception point position of the mesh center, the expansion parameter and the mesh tension function, is the target’s predicted future position;

[0183] The optimal tension calculation function F is a function based on the target equivalent size , a quadratic polynomial function of the target diagonal length is extracted from the intrinsic eigenvector. This function aims to balance the flexibility of the package with the rigidity of absorbing impact energy. The mathematical form is:

[0184]

[0185] in, It is the control parameter corresponding to the magnetic field strength required to be applied to the magnetorheological material network, and the unit is Tesla or Ampere. is the equivalent size of the target in meters; the coefficient It is an empirical constant obtained by conducting actual capture experiments on drones of different sizes and weights, measuring their impact kinetic energy and fitting them. For example, a typical set of coefficient values ​​is This calculation ensures that the mesh can achieve optimal capture performance for both small quadcopters and large multi-rotor drones.

[0186] : To achieve the best capture effect, the optimal tension that needs to be applied to the capture net is expressed in the form of control parameters:

[0187] : refers to the optimal tension calculation function; : target equivalent size parsed from intrinsic eigenvector; : Empirical coefficients of the quadratic polynomial function used to calculate the optimal tension;

[0188]

[0189] Example 7

[0190] A drone countermeasure method based on bionic intelligent capture includes the following steps:

[0191] Step 1: Multi-dimensional data perception: The data acquisition module's multi-source sensor array collects visible light, infrared, and radar echo signals from the target drone and performs weighted fusion to generate a comprehensive signal matrix that represents the target's multi-dimensional characteristics.

[0192] Step 2: State calculation and intent recognition: The integrated signal matrix is ​​processed using the falcon-like visual neural network model built into the state estimation module to calculate the target drone's current three-dimensional spatial position, speed, and other motion states, while simultaneously extracting its type, posture, and other intrinsic features.

[0193] Step 3: Future trajectory prediction: The motion state and intrinsic features of the historical time series are input into the long short-term memory network model of the trajectory prediction module to learn the dynamic behavior pattern of the target and generate its predicted position at the future moment;

[0194] Step 4: Countermeasure strategy generation. The countermeasure decision module makes a multimodal decision based on the target's predicted future position and intrinsic characteristics, combined with the intercepting drone's own flight status:

[0195] a) Determine the flight control strategy: Calculate the optimal flight modes and corresponding wingspan parameters, including cruise, tracking, and hovering, and generate the thrust and torque commands required for trajectory tracking;

[0196] b) Determine the capture net control strategy: evaluate the probability of successful capture and calculate the optimal instructions to trigger the capture net release and the net tension parameters required to ensure reliable capture;

[0197] Step 5: Integrated collaborative execution: Based on the generated strategy, the intercepting drone is precisely controlled to perform trajectory tracking and maneuvering. When the probability of successful capture meets the preset threshold, the capture net is deployed and the stiffness is adjusted to finally capture the target drone.

[0198] This embodiment describes the complete process of a drone countermeasure method based on bionic intelligent capture. The method begins with multidimensional data perception in step 1. The system's data acquisition module uses multi-source sensors to acquire signals and fuse them into a comprehensive signal matrix. It then proceeds to step 2: state solution and intent recognition. The state estimation module's falcon-like visual neural network model processes this matrix and outputs the target's real-time motion state and intrinsic characteristics. Based on this output, step 3: future trajectory prediction generates the target's future position using a long-short-term memory network model. In step 4: countermeasure strategy generation, the countermeasure decision module integrates all information and concurrently determines the flight control strategy and capture net control strategy, including selecting the optimal flight mode, adjusting wingspan, assessing capture probability, and setting net tension. In step 5: integrated collaborative execution, the system converts the generated complex instructions into precise control of the intercepting drone's flight dynamics and capture device, executing maneuvers and tracking. When the time is right, it implements release and tension adjustment, completing a complete, intelligent capture mission. The entire method forms a seamless closed loop from perception, understanding, prediction, decision-making, and action, demonstrating high adaptability and countermeasure efficiency against dynamic and unknown threats.

[0199] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A drone countermeasure system based on bionic intelligent capture, characterized by: include: The data acquisition module is used to obtain multi-dimensional feature data of the target UAV and generate a comprehensive signal matrix; a state estimation module, configured to receive the integrated signal matrix, calculate a motion state vector including target position and velocity, and extract an intrinsic feature vector including target type and posture; a trajectory prediction module, configured to generate a predicted future position of the target UAV based on the motion state vector and the intrinsic feature vector of the historical sequence; a countermeasure decision module, configured to generate a flight control strategy and a capture net control strategy based on the predicted future position, the intercepting drone's own flight state, and the intrinsic eigenvector; the flight control strategy determines the flight mode and wingspan parameters for trajectory tracking; and the capture net control strategy determines the release instruction and net tension parameters for triggering capture; When determining the delivery instruction, the countermeasure decision module is further configured to: Obtaining the predicted future position, and calculating a predicted interception accuracy factor, a relative speed factor, a target maneuverability factor, and a target attitude factor to form a capture evaluation index set; Constructing a capture probability evaluation model to calculate the probability of successful capture based on the capture evaluation indicator set; Comparing the success capture probability with a preset confidence threshold, and generating the delivery instruction when the success capture probability exceeds the confidence threshold; When determining the mesh tension parameter, the countermeasure decision module is further configured to: parsing equivalent size information of the target from the intrinsic feature vector generated by the state estimation module; An optimal tension calculation function is constructed to calculate the optimal mesh tension required to ensure reliable wrapping according to equivalent size information of the target; Setting the optimal mesh tension as the mesh tension parameter, wherein the mesh tension parameter is used to control the stiffness of the magnetorheological material mesh; The countermeasure decision module is further configured to: Obtaining the future predicted position and the mesh tension parameter; Constructing a mesh deployment trajectory optimization model, and inputting the mesh tension parameter as a physical constraint affecting the deployment behavior; Optimize the initial speed and duration of the deployment mechanism to ensure that the center of the deployed mesh can accurately intercept the predicted future position; When determining the flight control strategy, the countermeasure decision module is further configured to: Obtain target distance, relative speed, altitude difference and relative angle as flight status parameters; Constructing a mode switching decision function to calculate the fitness values ​​of six flight modes: cruising, tracking, diving, circling, hovering, and capturing based on the flight state parameters; Selecting the mode with the highest fitness value as the flight mode, wherein the flight mode is used for determining subsequent wingspan parameters; The countermeasure decision module is further configured to: Get the selected flight mode and the intercepting drone's own flight speed; Constructing a wingspan optimization function to dynamically fine-tune the basic wingspan corresponding to the flight mode and the flight speed of the intercepting drone itself; An optimal wingspan is calculated as the wingspan parameter, and the wingspan parameter is used to drive a dynamic wingspan adjustment mechanism to perform.

2. The drone countermeasure system based on bionic intelligent capture according to claim 1 is characterized in that: The data acquisition module is further used for: Collect visible light intensity signals, infrared intensity signals, and radar echo intensity signals obtained by visible light sensors, infrared sensors, and millimeter-wave radar sensors as multi-source perception data; The integrated signal matrix is ​​generated by performing weighted fusion on the multi-source perception data, and the integrated signal matrix is ​​used to provide input for the state estimation module.

3. The drone countermeasure system based on bionic intelligent capture according to claim 1 is characterized in that: The state estimation module is further configured to: Constructing a falcon-like visual neural network model, and using the integrated signal matrix as input of the falcon-like visual neural network model; Through the convolution and fully connected layer processing of the falcon-like visual neural network model, the motion state vector representing the three-dimensional spatial position and speed of the target is calculated, and the intrinsic feature vector representing the intrinsic properties of the target is extracted. The motion state vector and the intrinsic feature vector are used in the trajectory prediction module.

4. The drone countermeasure system based on bionic intelligent capture according to claim 1 is characterized in that: The trajectory prediction module is further configured to: Construct a long short-term memory network model; Inputting a feature sequence composed of the motion state vector at a historical moment and the intrinsic feature vector into the long short-term memory network model; The long short-term memory network model generates the future predicted position of the target UAV by learning the dynamic behavior of the target in consecutive frames. The future predicted position is used to provide a decision basis for the countermeasure decision module.

5. The drone countermeasure system based on bionic intelligent capture according to claim 1 is characterized in that: When determining the flight control strategy, the countermeasure decision module is further configured to: Obtaining the future predicted position generated by the trajectory prediction module and obtaining the current flight status of the intercepting UAV; Constructing a trajectory tracking control law to calculate the total thrust and torque vector required to achieve tracking of the predicted future position; The total thrust and torque vector is decomposed into thrust instructions for each independent fan, and the thrust instructions are used to execute the flight control strategy.

6. A drone countermeasure method based on bionic intelligent capture, using the drone countermeasure system based on bionic intelligent capture according to any one of claims 1 to 5, characterized in that: The following steps are involved: Step 1: Multi-dimensional data perception: The multi-source sensor array of the data acquisition module collects the visible light, infrared and radar echo signals of the target UAV and performs weighted fusion to generate a comprehensive signal matrix that represents the multi-dimensional characteristics of the target; Step 2: State calculation and intention recognition: The integrated signal matrix is ​​processed using the falcon-like visual neural network model built into the state estimation module to calculate the current three-dimensional spatial position and velocity state of the target UAV, and simultaneously extract its type and posture intrinsic features; Step 3: Future trajectory prediction: The motion state and intrinsic features of the historical time series are input into the long short-term memory network model of the trajectory prediction module to learn the dynamic behavior pattern of the target and generate its predicted position at the future moment through the long short-term memory network model; Step 4: Countermeasure strategy generation. The countermeasure decision module makes a multimodal decision based on the target's predicted future position and intrinsic characteristics, combined with the intercepting drone's own flight status: a) Determine the flight control strategy: Calculate the fitness values ​​of the six flight modes (cruise, track, dive, hover, hover, and capture) based on the flight state parameters, and select the flight mode with the highest fitness value; b) Determine the capture net control strategy: evaluate the probability of successful capture and calculate the optimal instructions to trigger the capture net release and the net tension parameters required to ensure reliable capture; Step 5: Integrated collaborative execution: Based on the generated strategy, the intercepting UAV is precisely controlled to perform trajectory tracking and maneuvering. When the probability of successful capture meets the preset threshold, the release and stiffness adjustment of the capture net are triggered, and the target UAV is finally captured.

Images