Unmanned aerial vehicle cluster collaborative target hunting method based on brain-like calculation

Abstract

The invention discloses an unmanned aerial vehicle cluster collaborative target hunting method based on brain-like calculation, and belongs to the technical field of autonomous navigation robot and multi-robot coordination control, and the method comprises the steps: obtaining a multi-view image under a camera view, and obtaining the local pose and covariance of each unmanned aerial vehicle; obtaining a global consistent relative pose of the unmanned aerial vehicle cluster according to the local pose and the covariance of each unmanned aerial vehicle; performing state estimation on the target by adopting an unscented Kalman filtering algorithm to obtain target state information under a unified coordinate system; using a Bezier curve generation algorithm to obtain a target motion trail in a section of historical state; the method comprises the following steps: expanding an unmanned aerial vehicle cluster through a centroid extension method to obtain a surrounding queue of target surrounding, solving the minimum surrounding cost by adopting a Gaussian Newton method, and generating a surrounding position of each unmanned aerial vehicle; the motion trajectory of the unmanned aerial vehicle is generated by adopting mixed A * search, and a final surrounding trajectory sequence is generated, so that the accuracy of motion state estimation is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous navigation robots and multi-robot coordinated control technology, specifically relating to a method for collaborative target capture of unmanned aerial vehicle (UAV) swarms based on brain-like computing. Background Technology

[0002] With the profound evolution of information technology, the battlefield situation is characterized by high dynamism and intense confrontation. Against this backdrop, the rapid identification, continuous tracking, and precise acquisition of high-speed maneuvering targets have become core capabilities for improving tactical reconnaissance effectiveness and building proactive defense systems. Unmanned aerial vehicle (UAV) swarms, with their distributed collaborative advantages, powerful spatial coverage capabilities, and flexible formation reorganization characteristics, offer a revolutionary solution for target acquisition missions in complex battlefield environments. Traditional UAV swarm collaborative acquisition models are typically based on two key assumptions: first, reliance on global absolute position information provided by the Global Positioning System (GPS) or similar infrastructure; and second, the relatively stable or predictable movement patterns of the targets, with the acquisition process primarily employing pre-defined fixed geometric formations (such as spherical encirclement or cone interception). These methods perform well in open areas and low-speed target scenarios. However, when the operational environment evolves into complex urban canyons, jungles, or underground spaces where GPS signals are scarce and obstacles are dense, and the targets are high-speed, highly flexible, maneuverable, and non-cooperative objects, the limitations of traditional models are significantly amplified, even leading to systemic failures.

[0003] Specifically, the complex operating environment and highly dynamic targets pose significant challenges to the cooperative capture of UAV swarms. In environments without a Global Positioning System (GPS), UAV swarms cannot obtain precise global coordinates of themselves or their companions, nor can they accurately anchor the absolute positions of targets and obstacles. Therefore, cooperative control strategies lack spatial reference, severely restricting the sharing of contextual awareness and decision-making consistency across large-scale UAV swarms. Furthermore, observations from individual UAVs based on local sensors are susceptible to occlusion and noise interference, leading to significant biases and poor robustness in estimation results. Traditional algorithms also experience a significant reduction in accuracy when dealing with highly nonlinear motion, resulting in misjudgments of target position, velocity, and even intent. To establish global consistency between the UAV swarm and the target, simply fusing observation data from multiple UAVs faces problems such as spatiotemporal misalignment and observation heterogeneity. Finally, to rapidly surround and capture the target, the UAV swarm must simultaneously meet the following conditions: rapidly forming an effective encirclement, strictly avoiding obstacles, preventing intra-swarm collisions, and maintaining target visibility, among others. These conditions are often conflicting constraints.

[0004] Traditional methods often employ incremental optimization or rule bases, making it difficult to strike a balance between real-time performance and safety. This can easily lead to drones getting stuck in a decision-making deadlock state of "oscillating hesitancy" or "path blocking." Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a brain-inspired computing-based UAV swarm cooperative target capture method that solves the problem that existing related methods suffer from complex operating environments and highly dynamic target cooperative capture, making it difficult to achieve a balance between real-time performance and security.

[0006] To achieve the aforementioned objectives, the technical solution adopted by this invention is as follows: a method for collaborative target capture by a swarm of unmanned aerial vehicles (UAVs) based on neuromorphic computing, comprising the following steps: S1. Based on the multi-view images obtained from the camera perspective of the UAV, the local pose and covariance of each UAV are obtained by using panoramic images, visual motion and grid-landmark cell models. S2. Based on the local pose and covariance of each UAV, the globally consistent relative pose of the UAV swarm is obtained through compressed broadcasting, common-view matching and graph optimization algorithms. S3. Based on the globally consistent relative pose of the UAV swarm, each UAV identifies the target, uses the unscented Kalman filter algorithm to estimate the target state, and fuses all the target state estimates to obtain the target state information in a unified coordinate system. S4. Based on historical target state information, use the Bézier curve generation algorithm to obtain a target motion trajectory under a historical state, expand the Bézier curve trajectory to predict the target motion, and obtain the target prediction state under a unified coordinate system. S5. Expand the drone swarm by extending the centroid to obtain the target capture queue. Use the Gauss-Newton method to solve for the minimum capture cost, and then obtain the optimal capture queue to generate the capture position of each drone. S6. Based on the capture position of each drone, a hybrid A* search algorithm is used to generate the drone's motion trajectory. Collision avoidance detection is performed based on the motion trajectory to generate the final capture trajectory sequence, thus completing the capture of the agile target.

[0007] Furthermore, S1 includes the following sub-steps: S11. Based on the multi-view images obtained by the UAV cluster through the monocular camera, a continuous panoramic image is generated using the back projection and spherical mapping algorithm. S12. Based on the panoramic image, the ORB operator and RANSAC five-point method are used to obtain the rotation matrix and translation vector between adjacent frames, which are used as the measurement of visual motion. S13. Based on the feature description amplitude, the first pulse coding and grid-landmark cell model are used to perform position coding and memory; S14. Based on visual motion, IMU data and position encoding, UKF is used to predict and correct the state, and output the local pose and covariance of each UAV.

[0008] Furthermore: In S11, the method of using back projection and spherical mapping algorithm specifically involves back-projecting the multi-view image onto the body coordinate system, wherein, for the first... i drone camera pixels The specific expression for back projection is: In the formula, These are the pixel coordinates projected onto the UAV's body coordinate system. Let the rotation matrix of the UAV in the world coordinate system be denoted as . The translation matrix of the starting point of the UAV in the world coordinate system. For the first i The intrinsic parameter matrix of the drone camera, Y i The coordinates are in the body coordinate system. u Let be the coordinates of the image in the horizontal direction. v The coordinates of the image in the vertical direction; All Y i In radius Mapped onto the angular space on the sphere Panoramic images were synthesized using bilinear interpolation. ; S12 specifically involves: extracting key points and their descriptors from the panoramic image using the ORB operator, pairing them using FLANN, eliminating outliers using RANSAC, and then using the five-point method to calculate the relative motion under known intrinsic parameters to obtain the measured visual motion. ; In the formula, Let be a rotation matrix. It is a translation vector; In S13, the grid-landmark cell model includes grid cells and landmark cell networks. Specifically, S13 maps feature description amplitudes to pulse times. , generating sparse pulses; In the formula, To prevent small amounts from being divided by zero, This is the upper limit of the encoding time window. For descriptors, The descriptor for the maximum amplitude at the current moment; When sparse impulses are input into the grid cell and landmark cell network, the grid cell state vector is updated using the following formula; In the formula, It is a non-linear activation. and For learnable parameters, For time t The grid cell state vector, For time t +1 grid cell state vector; Landmark cells activate associated memories by activating sparse pulse and grid cell state vectors according to the Hebbian rule; In the formula, To activate associated memories, To activate memories prior to association, For time t sparse pulses, For learning rate, It is the transpose symbol; In subsequent frames, correction parameters are established based on the memory associated with activation. Correct the grid code, and use the corrected grid code as the position code; In the formula, For learning rate, For time t +1 sparse pulse; In S14, the state vector of UKF and observation for: In the formula, For time t The local pose, For time t speed; The first estimate was obtained through UKF nonlinear prediction. Compared with the second forecast The Kalman gain is calculated based on the second prediction. ; In the formula, H The state transition equation is... This is the noise matrix; The UKF state vector is updated based on the Kalman gain and the first prediction, and the local pose and covariance of each UAV are output through the UKF. In the formula, This is the updated state vector from UKF. This is a transformation of the state transition equation.

[0009] Furthermore, S2 includes the following sub-steps: S21. Based on covariance and sparse impulses, differential quantization or autoencoder algorithms are used to generate broadcast data packets for sharing neighboring machine observations. S22. Estimate the overlapping area of ​​the field of view based on the broadcast data packets, and use the ORB+RANSAC algorithm to calculate the relative observations between UAVs; S23. Based on the relative observations between UAVs, the incremental Gauss-Newton algorithm is used to solve the global pose and covariance of the UAVs. S24. Send the global pose and covariance of the UAV to each UAV, perform closed-loop correction of the global pose, and obtain the globally consistent relative pose of the UAV cluster.

[0010] Furthermore, S21 specifically involves: performing differential quantization on the output local pose increment and covariance of each UAV to obtain the differentially quantized parameters. and : In the formula, A unified quantization function for vectors and matrices. For time t covariance, For time t +1 covariance, For local pose increments; In the formula, For time t -1 local pose; A sparse autoencoder is used to reduce the dimensionality of the grid cell state vector and sparse impulses to obtain the dimensionality reduction result. h ; In the formula, To concatenate vectors, For activation function, For sparse autoencoder parameters, State vector The corresponding estimated value; Establish broadcast data packets Broadcast data packets within the maximum latency allowed by link quality Reaching adjacent drones within the range; S22 specifically involves: obtaining the neighboring machine's grid activation by receiving broadcast data packets from the neighboring machine; calculating the field of view center deviation based on the UAV's own grid activation and the neighboring machine's grid activation; estimating the center and radius of the overlapping field of view circle; and obtaining the overlapping field of view area. Among them, the field of view center deviation includes the first spherical coordinate deviation. Second spherical coordinate deviation ; In the formula, Activate the drone's own mesh. Activate the neighboring machine's mesh. and This is a mapping function from grid encoding to spherical coordinate differences; ORB operators are used to extract segments of the decoded panoramic image in the overlapping regions of the field of view, and joint descriptors are constructed by combining the temporal information of the first pulse encoding. Matching pairs are obtained after RANSAC filtering. In the known camera intrinsic parameters K Solving the relative observations between UAVs under certain conditions This includes the rotation matrix of the drone itself relative to its neighboring drones. Translation vector ; In the formula, R For the first i The drone and the first j The difference in rotational variation between the drones For the first i A drone in the camera's internal reference K In this case, the difference in translation from the initial position, For the first j A drone in the camera's internal reference K In this case, the difference in translation from the initial position; Observation covariance of relative observations between UAVs The specific expression is: In the formula, The number of all drones to establish contact; S23 specifically involves: establishing a graph optimization model for a drone swarm, where the node set represents the local pose of each drone, the edge set consists of the relative observations and observation covariance among the drones, and the objective function of the graph optimization model is... The specific expression is: In the formula, For the first in three-dimensional space j The difference in pose between the drone and its initial position. For the first in three-dimensional space i The difference in pose between the drone and its initial position. for and The pose difference between them For SE(3) complex; The incremental Gauss-Newton algorithm works as follows: upon receiving new relative observations, it linearizes the residuals. The sparse information matrix is ​​incrementally updated, and the local pose of the UAV is updated after the increment is solved, which serves as the global pose of the UAV. In the formula, For the objective function The first partial derivative, For the objective function The first differential, For the objective function The second partial derivative, For the objective function The second differential, This is the difference between the sum of the partial derivative and the differential product; In the formula, The sparse information matrix before the update. This is the updated sparse information matrix; In the formula, p i This is the local pose of the drone before the update. This represents the updated local pose of the drone. S24 specifically involves broadcasting the global pose and covariance of the UAV to each UAV, and updating the UKF state of the local UAV as follows: In the formula, The updated state vector, This represents the global pose of the drone. This is the updated grid cell state vector. P i For covariance, and For the prior noise of velocity and mesh activation, To update the pose of the drone in three-dimensional space. For the updated speed; Various drones utilize and Restart the filtering loop and input the updated grid cell state vector and sparse impulses into the landmark cells for long-term memory updates to obtain the globally consistent relative pose of the UAV cluster.

[0011] Furthermore, S3 includes the following sub-steps: S31. Using the YOLO target recognition method, identify the target in the image information and obtain the target position in its coordinate system based on the image information; S32. Transform the target position using a rotation matrix to obtain the target position in a unified coordinate system; S33. Based on the target position in a unified coordinate system, UKF is used to estimate the target state, and the target state estimate and covariance after filtering out sensor errors are obtained. S34. Using a loosely coupled approach, the noise covariance of each UAV's observation of the target is combined to fuse all target state estimates and obtain target state information. S33 specifically refers to: Based on the state vector using UKF Covariance Matrix Generate 2 n The formula for generating +1 sigma point is as follows: In the formula, For sigma points, n Let be the dimension of the state vector. for K The estimated value corresponding to the position at a given time. For adjustment functions; In the formula, and To adjust the parameters; The predicted state mean is calculated by propagating the sigma point through a nonlinear state transition function. Covariance ; In the formula, To pass K The state value at time t is predicted. K The value at time +1, Q To measure the covariance matrix of the noise, , For the state part, For the noise part, and These are the weighting coefficients; In the formula, It is a tiny error; Map the predicted Sigma points to the measurement space and calculate the mean of the measurements. Covariance matrix S and cross-covariance matrix T ; In the formula, To pass K The observed values ​​at time 10 are predicted. K The value at time +1, R i This is the assumed pre-defined noise covariance matrix; Calculate the Kalman gain based on the covariance matrix. ; Based on the observed values and the mean of the measurements Update the target state estimate and covariance; updated target state estimate Covariance The expression is: In the formula, for K The observed value at time; In S34, the target state information is obtained. The specific expression is: In the formula, For the first i Target state estimation of the drone For the first The weight of the drone; In the formula, For the first The observation noise covariance of the drone.

[0012] Furthermore, S4 includes the following sub-steps: S41. Use the sliding window method to record historical target status information of a set length; S42. Based on historical target state information, generate a target motion trajectory under historical state according to the Bézier curve generation algorithm; S43. Calculate the predicted step size using the hyperbolic tangent function based on the distance between the centroid of the UAV cluster and the current target position. S44. Extend the predicted step size of the Bézier curve to obtain the predicted state of the target under a unified coordinate system. S42 specifically refers to: using Bézier curves to describe the target's trajectory, the target's trajectory... The specific expression is: In the formula, Given an nth-degree Bernstein polynomial basis, These are the control points for the Bézier curve; In S43, the prediction step size is calculated. The specific expression is: In the formula, For the pre-set prediction time step, d The distance to the target's current location. This is the current time.

[0013] Furthermore, S5 includes the following sub-steps: S51. Calculate the centroid of the UAV swarm based on the globally unified coordinate system. S52. Calculate the vector from the centroid to the drone in each drone. All drones expand outward to form a capture queue. Sort the drones in the centroid region and the small fan-shaped region by permutation and combination to form several capture queues. S53. Use the Gauss-Newton method to solve the capture function and obtain the minimum capture cost corresponding to the capture queue. S54. Compare the minimum capture cost corresponding to all candidate capture queues, take the capture queue with the minimum minimum capture cost as the optimal capture queue, calculate the optimal capture point, and thus obtain the capture position of each drone. S52 specifically involves: obtaining the vector of each drone relative to its centroid, calculating the angle between this vector and the drone's encirclement direction vector, where the calculation of the... i The angle of the drone The specific expression is: In the formula, It is a modulo function. For vectors and x The included angle between the axes, with values ​​ranging from [0, 2π). For the first i The vector of the drone relative to its center of mass. , C Center of mass, , N For the number of drones, The direction vector for encirclement and capture. , T The target encirclement point; Sort all the drones by their included angles, assign serial numbers to the drones based on the sorting results, and generate a preliminary encirclement order. If there is a drone at the centroid, insert the drone into the circular formation between two adjacent drones in sequence, and generate an encirclement queue according to all possible insertion orders. If there are more than a set number of drones in the same direction, and the angle between the dense drones in the same sector is less than a preset threshold, then the drone set is arranged and combined to obtain all possible insertion orders, and inserted into the interval position of the initial encirclement order. An encirclement queue is generated according to all possible insertion order. In S53, the minimum capture cost corresponding to the capture queue is calculated. The specific expression is: In the formula, S For the encirclement and capture queue, For the initial angle offset, R The radius of the encirclement. The cost function; In the formula, For the encirclement queue S The Middle j The actual location of the drone.

[0014] Furthermore, S6 includes the following sub-steps: S61. Obtain information about surrounding obstacles from the binocular camera, construct a three-dimensional grid or octree environment model, and map the continuous state of the UAV to discrete grid points and discrete heading angles for searching on the grid map. The continuous state includes position, orientation and speed. S62. Using the current position, orientation, and speed of the UAV as the starting point for the search, determine the safe encirclement point and attitude of the target area or target group according to the predetermined "capture and encircle" strategy. S63. In the discretized state space, the kinematic constraints of the UAV are used as the extended rules. A heuristic function is used to evaluate the cost from the current grid point to the target, which drives the search towards the target and generates a connected trajectory that satisfies the dynamic constraints. S64. Perform spline fitting or smoothing processing to minimize bending energy on the discrete trajectory generated by the hybrid state A algorithm to eliminate sharp turns in the broken line segment. Based on the speed and acceleration limits of the UAV, assign speed and timestamp to the smooth trajectory to obtain a continuous and executable time-parameterized trajectory. S65. Discretize the smooth trajectory and use the nearest obstacle distance detection algorithm to detect obstacle collisions at each sampling point. In response to the detection of a collision or insufficient safety distance, trigger local replanning and perform high and low altitude operations near that segment to correct the path in the smooth trajectory. S66. Integrate and splice the smooth trajectory, the time-parameterized trajectory, and the trajectory segments that have passed collision detection to generate the final capture trajectory sequence; S64 specifically refers to: Goal-oriented kinematic search front-end: Constructs a search tree based on the hybrid state A algorithm, generates motion primitives through discretized control input, and introduces a cost function during the search process. Using the target trajectory B(t) as heuristic information, a dual heuristic function is designed: In the formula, For dynamic distance estimation based on the optimal boundary value problem, For the drone target status, This is the current status of the drone. As weight, S t For the sum of the expected extended time, For path unfolding time, The following is obtained by extrapolating the predicted trajectory: In the formula, The Bézier curve trajectory of the original historical trajectory. The trajectory obtained by extrapolation over time; Spatiotemporal trajectory optimization backend: Based on the initial path generated by the goal-oriented kinematic search frontend, a flight corridor composed of connected free-space cubes is constructed. ; In the formula, These are the control points of the Bézier curve. M The number of trajectory segments in the flight corridor; Establish an optimization problem with dynamic constraints under corridor constraints: In the formula, For smoothing terms, For the field of power, For dynamic constraint terms, q w intermediate waypoints T For the time of segmented trajectory, for x , y and z The combination x For segmented waypoints x Axis coordinates y For segmented waypoints y Axis coordinates z For segmented waypoints z Axis coordinates; This optimization problem is solved efficiently using the split convex optimization method, generating a time-parameterized trajectory; In S66, the collision detection method is as follows: For each UAV's time-parameterized trajectory, let the UAV... The trajectory is m Composition of segmental polynomials, defining the parameter set ; In the formula, For the first k Duration of segment , For the real number field, For the first k Segment polynomial order , For the matrix field, The coefficient matrix, , For the first k segment polynomial x Parameters in the axial direction, For the first k segment polynomial y Parameters in the axial direction, For the first k segment polynomial z Parameters in the axial direction; Calculate the relative time of the current moment within each segment using a polynomial. ; In the formula, It is the difference between the current time and the starting time. For absolute time, This is the starting timestamp of the trajectory. It is a function with maximum value. s The number of trajectory segments, For the minimum value index function, The time for each trajectory segment; Calculate the three-dimensional spatial position of the predicted trajectory ; In the formula, This is the current location of the drone. For the first s Duan Di j Term coefficient, , for x Coefficient in the axial direction, for y Coefficient in the axial direction, for z Coefficient in the axial direction, In the first s Relative time within a segment; drones The sampling point at each moment in the predicted trajectory is compared with the position of the predicted trajectory of other drones at the same moment. If the distance is lower than the preset safe distance, it is determined that there is a potential collision risk. When a potential collision risk is identified, each drone with this risk receives the target's status information in real time and simultaneously acquires the position information of other drones. Each drone calculates its own Euclidean distance to the target and compares it with the Euclidean distances of other drones to the target. Drones are then sorted in ascending order based on their Euclidean distances to determine their movement priority. The drone with the highest priority maintains its planned trajectory, while the remaining drones perform obstacle avoidance maneuvers at different altitudes according to their priority ranking. The specific expression is: In the formula, For the first j The location coordinates of the drone Here are the target location coordinates, and rank(•) is the sorting function. The beneficial effects of this invention are as follows: (1) The method of the present invention solves the problem of relative positioning of UAV swarms in denied environments. It constructs a globally unified positioning through swarm collaboration and global optimization, thereby laying the foundation for collaborative operation. The UAV swarm can effectively cope with environmental uncertainties and its own mobility.

[0015] (2) The method of the present invention has made changes to the collaborative target state estimation and prediction algorithm. It integrates the unscented Kalman filter algorithm and the Bezier curve generation algorithm in the trajectory prediction method of multi-source observation, which significantly improves the accuracy of motion state estimation.

[0016] (3) The method of this invention has made changes to the encirclement strategy algorithm. It innovatively proposes an optimization strategy generation algorithm based on the centroid extension of the virtual centroid, as well as a spatiotemporal trajectory optimization method, so as to achieve rapid target encirclement under the premise of ensuring safety and generate a dynamic feasible trajectory. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for collaborative target capture by a drone swarm based on brain-like computing, according to the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0019] like Figure 1 As shown, in one embodiment of the present invention, a method for cooperative target capture of a drone swarm based on neuromorphic computing includes the following steps: S1. Based on the multi-view images obtained from the camera perspective of the UAV, the local pose and covariance of each UAV are obtained by using panoramic images, visual motion and grid-landmark cell models. S2. Based on the local pose and covariance of each UAV, the globally consistent relative pose of the UAV swarm is obtained through compressed broadcasting, common-view matching and graph optimization algorithms. S3. Based on the globally consistent relative pose of the UAV swarm, each UAV identifies the target, uses the unscented Kalman filter algorithm to estimate the target state, and fuses all the target state estimates to obtain the target state information in a unified coordinate system. S4. Based on historical target state information, use the Bézier curve generation algorithm to obtain a target motion trajectory under a historical state, expand the Bézier curve trajectory to predict the target motion, and obtain the target prediction state under a unified coordinate system. S5. Expand the drone swarm by extending the centroid to obtain the target capture queue. Use the Gauss-Newton method to solve for the minimum capture cost, and then obtain the optimal capture queue to generate the capture position of each drone. S6. Based on the capture position of each drone, a hybrid A* search algorithm is used to generate the drone's motion trajectory. Collision avoidance detection is performed based on the motion trajectory to generate the final capture trajectory sequence, thus completing the capture of the agile target.

[0020] S1 includes the following steps: S11. Based on the multi-view images obtained by the UAV cluster through the monocular camera, a continuous panoramic image is generated using the back projection and spherical mapping algorithm. S12. Based on the panoramic image, the ORB operator and RANSAC five-point method are used to obtain the rotation matrix and translation vector between adjacent frames, which are used as the measurement of visual motion. S13. Based on the feature description amplitude, the first pulse coding and grid-landmark cell model are used to perform position coding and memory; S14. Based on visual motion, IMU data and position encoding, the UKF (Unscented Kalman Filter) algorithm is used to predict and correct the state, and output the local pose and covariance of each UAV.

[0021] In S11, the method using back projection and spherical mapping algorithm is as follows: the multi-view image is back-projected onto the body coordinate system, wherein, for the ... i drone camera pixels The specific expression for back projection is: In the formula, These are the pixel coordinates projected onto the UAV's body coordinate system. Let the rotation matrix of the UAV in the world coordinate system be denoted as . This is the translation matrix of the UAV at its starting point (origin) in the world coordinate system. For the first i The intrinsic parameter matrix of the drone camera, Y i The coordinates are in the body coordinate system. u Let be the coordinates of the image in the horizontal direction. v The coordinates of the image in the vertical direction; All Y i In radius Mapped onto the angular space on the sphere Panoramic images were synthesized using bilinear interpolation. In this embodiment, the process of synthesizing the panoramic image maintains the geometric consistency of the views from each camera, providing a comprehensive environmental observation for subsequent visual processing.

[0022] S12 specifically involves: extracting key points and their descriptors from the panoramic image using the ORB operator, pairing them using FLANN, eliminating outliers using RANSAC, and then using the five-point method to calculate the relative motion under known intrinsic parameters to obtain the measured visual motion. ; In the formula, Let be a rotation matrix. It is a translation vector; In S13, the grid-landmark cell model includes grid cells and landmark cell networks. Specifically, S13 maps feature description amplitudes to pulse times. , generating sparse pulses; In the formula, To prevent small amounts from being divided by zero, This is the upper limit of the encoding time window. For descriptors, The descriptor for the maximum amplitude at the current moment; When sparse impulses are input into the grid cell and landmark cell network, the grid cell state vector is updated using the following formula; In the formula, It is a non-linear activation. and For learnable parameters, For time t The grid cell state vector, For time t +1 grid cell state vector; Landmark cells activate associated memories by activating sparse pulse and grid cell state vectors according to the Hebbian rule; In the formula, To activate associated memories, To activate memories prior to association, For time t sparse pulses, The learning rate; In subsequent frames, correction parameters are established based on the memory associated with activation. Correct the grid code, and use the corrected grid code as the position code; In the formula, It is the transpose symbol. For learning rate, For time t +1 sparse pulse; In S14, the state vector of UKF and observation for: In the formula, For time t The local pose, For time t speed; The first estimate was obtained through UKF nonlinear prediction. Compared with the second forecast The Kalman gain is calculated based on the second prediction. ; In the formula, H The state transition equation is... This is the noise matrix; The UKF state vector is updated based on the Kalman gain and the first prediction, and the local pose and covariance of each UAV are output through the UKF. In the formula, This is the updated state vector from UKF. This is a transformation of the state transition equation.

[0023] In this embodiment, the final output is the local pose. With covariance This provides a reliable input for cluster integration.

[0024] S2 includes the following steps: S21. Based on covariance and sparse impulses, differential quantization or autoencoder algorithms are used to generate broadcast data packets for sharing neighboring machine observations. S22. Estimate the overlapping area of ​​the field of view based on the broadcast data packets, and use the ORB+RANSAC algorithm to calculate the relative observations between UAVs; S23. Based on the relative observation between UAVs, the incremental Gauss-Newton algorithm is used to solve the global pose and covariance of the UAVs, and the global pose is solved in real time. S24. The global pose and covariance of the UAV are sent to each UAV, and the global pose is corrected in a closed loop to obtain the globally consistent relative pose of the UAV cluster, thereby improving the closed-loop accuracy.

[0025] S21 specifically involves: performing differential quantization on the output local pose increment and covariance of each UAV to obtain the differentially quantized parameters. and : In the formula, A unified quantization function for vectors and matrices. For time t covariance, For time t +1 covariance, For local pose increments; In the formula, For time t -1 local pose; In this embodiment, to reduce bandwidth overhead, each UAV performs differential quantization on its output local pose increment and covariance.

[0026] A sparse autoencoder is used to reduce the dimensionality of the grid cell state vector and sparse impulses to obtain the dimensionality reduction result. h ; In the formula, To concatenate vectors, For activation function, For sparse autoencoder parameters, State vector The corresponding estimated value; Establish broadcast data packets Broadcast data packets within the maximum latency allowed by link quality Reaching adjacent drones within the range; S22 specifically involves: obtaining the neighboring machine's grid activation by receiving broadcast data packets from the neighboring machine; calculating the field of view center deviation based on the UAV's own grid activation and the neighboring machine's grid activation; estimating the center and radius of the overlapping field of view circle; and obtaining the overlapping field of view area. Among them, the field of view center deviation includes the first spherical coordinate deviation. Second spherical coordinate deviation ; In the formula, Activate the drone's own mesh. Activate the neighboring machine's mesh. and This is a mapping function from grid encoding to spherical coordinate difference, used to estimate the center and radius of the overlapping field of view, thereby obtaining the overlapping area of ​​the field of view; ORB operators are used to extract segments of the decoded panoramic image in the overlapping regions of the field of view, and joint descriptors are constructed by combining the temporal information of the first pulse encoding. Matching pairs are obtained after RANSAC filtering. In the known camera intrinsic parameters K Solving the relative observations between UAVs under certain conditions This includes the rotation matrix of the drone itself relative to its neighboring drones. Translation vector ; In the formula, R For the first i The drone and the first j The difference in rotational variation between the drones For the first i A drone in the camera's internal reference K In this case, the difference in translation from the initial position, For the first j A drone in the camera's internal reference K In this case, the difference in translation from the initial position; Observation covariance of relative observations between UAVs The specific expression is: In the formula, The number of all drones to establish contact; S23 specifically involves: establishing a graph optimization model for a drone swarm, where the node set represents the local pose of each drone, the edge set consists of the relative observations and observation covariance among the drones, and the objective function of the graph optimization model is... The specific expression is: In the formula, For the first in three-dimensional space j The difference in pose between the drone and its initial position. For the first in three-dimensional space i The difference in pose between the drone and its initial position. for and The pose difference between them For SE(3) complex; The incremental Gauss-Newton algorithm works as follows: upon receiving new relative observations, it linearizes the residuals. And incrementally update the sparse information matrix, and solve the incremental problem. The local pose of the drone is then updated as the global pose of the drone. In the formula, For the objective function The first partial derivative, For the objective function The first differential, For the objective function The second partial derivative, For the objective function The second differential, This is the difference between the sum of the partial derivative and the differential product; In the formula, The sparse information matrix before the update. This is the updated sparse information matrix; In the formula, p i This is the local pose of the drone before the update. This represents the updated local pose of the drone. S24 specifically involves broadcasting the global pose and covariance of the UAV to each UAV, and updating the UKF state of the local UAV as follows: In the formula, The updated state vector, This represents the global pose of the drone. This is the updated grid cell state vector. P i For covariance, and For the prior noise of velocity and mesh activation, To update the pose of the drone in three-dimensional space. For the updated speed; Various drones utilize and Restart the filtering loop and input the updated grid cell state vector and sparse impulses into the landmark cells for long-term memory updates to obtain the globally consistent relative pose of the UAV cluster.

[0027] Through the above closed-loop correction, the system tightly couples the local and global solutions in large-scale formations, achieving high-precision, low-drift end-to-end relative positioning.

[0028] S3 includes the following steps: S31. Using the YOLO target recognition method, identify the target in the image information and obtain the target position in its coordinate system based on the image information; S32. Transform the target position using a rotation matrix to obtain the target position in a unified coordinate system; S33. Based on the target position in a unified coordinate system, UKF is used to estimate the target state, and the target state estimate and covariance after filtering out sensor errors are obtained. S34. Using a loosely coupled approach, the noise covariance of each UAV's observation of the target is combined to fuse all target state estimates and obtain target state information. S33 specifically refers to: Based on the state vector using UKF Covariance Matrix Generate 2 n The formula for generating +1 sigma point is as follows: In the formula, For sigma points, n Let be the dimension of the state vector. for K The estimated value corresponding to the position at a given time. For adjustment functions; In the formula, and To adjust the parameters; The predicted state mean is calculated by propagating the sigma point through a nonlinear state transition function. Covariance ; In the formula, To pass K The state value at time t is predicted. K The value at time +1, Q To measure the covariance matrix of the noise, , For the state part, For the noise part, and These are the weighting coefficients; In the formula, It is a tiny error; Map the predicted Sigma points to the measurement space and calculate the mean of the measurements. Covariance matrix S and cross-covariance matrix T The cross-covariance matrix is ​​calculated to determine the accuracy of the calculated measurement values; In the formula, To pass K The observed values ​​at time 10 are predicted. K The value at time +1, R i This is the assumed pre-defined noise covariance matrix; Calculate the Kalman gain based on the covariance matrix. ; Based on the observed values and the mean of the measurements Update the target state estimate and covariance; updated target state estimate Covariance The expression is: In the formula, for K The observed value at time; In S34, the target state information is obtained. The specific expression is: In the formula, For the first i Target state estimation of the drone For the first The weight of the drone; In the formula, For the first The observation noise covariance of the drone.

[0029] In this embodiment, multiple UAVs observe the target's position, and a weighted average method is used to fuse the observation results from each UAV to obtain the target's state information. This method enables rapid sharing of accurate target state information among the cluster.

[0030] S4 includes the following steps: S41. Use the sliding window method to record historical target status information of a set length; S42. Based on historical target state information, generate a target motion trajectory under historical state according to the Bézier curve generation algorithm; S43. Calculate the predicted step size using the hyperbolic tangent function based on the distance between the centroid of the UAV cluster and the current target position. S44. Extend the predicted step size of the Bézier curve to obtain the predicted state of the target under a unified coordinate system. S42 specifically refers to: using Bézier curves to describe the target's trajectory, the target's trajectory... The specific expression is: In the formula, Given an nth-degree Bernstein polynomial basis, These are the control points for the Bézier curve; In this embodiment, the target trajectory is described using Bernstein basis polynomials, referred to as B'ezier curves. This invention uses t∈ The 3D position of the target observed in the global frame at time step is denoted as . Then, maintain a FIFO queue of length L to store past observations and their corresponding timestamps. This queue is represented as... =[ , ,…, ],in ={ , } The included time range is [ , ],in It equals the current time. When new target observations are obtained, a new target trajectory is also generated by fitting past observations.

[0031] In S43, the prediction step size is calculated. The specific expression is: In the formula, For the pre-set prediction time step, d The distance to the target's current location. This is the current time.

[0032] In this embodiment, in order to quickly capture unintentional targets during the capture process, the present invention uses the hyperbolic tangent function tanh(x) to determine the predicted target position when the distance d between the centroid and the target's current position is different.

[0033] S5 includes the following steps: S51. Calculate the centroid of the UAV swarm based on the globally unified coordinate system. S52. Calculate the vector from the centroid to the drone in each drone. All drones expand outward to form a capture queue. Sort the drones in the centroid region and the small fan-shaped region by permutation and combination to form several capture queues. S53. Use the Gauss-Newton method to solve the capture function and obtain the minimum capture cost corresponding to the capture queue. S54. Compare the minimum capture cost corresponding to all candidate capture queues, take the capture queue with the minimum minimum capture cost as the optimal capture queue, calculate the optimal capture point, and thus obtain the capture position of each drone. In S51, at time t, the positions of the drone swarm are known, and the position of each drone is... The centroid of the cluster , N Given the number of drones, the target capture point is set as... T Then, the direction from the centroid C to the target encirclement point... T A capture direction vector can be obtained: .

[0034] S52 specifically involves: obtaining the vector of each drone relative to its centroid, calculating the angle between this vector and the drone's encirclement direction vector, where the calculation of the... i The angle of the drone The specific expression is: In the formula, It is a modulo function. For vectors and x The included angle between the axes, with values ​​ranging from [0, 2π). For the first i The vector of the drone relative to its center of mass. , C Center of mass; Sort all the drones by their included angles, assign serial numbers to the drones based on the sorting results, and generate a preliminary encirclement order. In this embodiment, let , where θ( ) represents a vector and x The included angle of the axes, ranging from [0, 2π); here they are sorted clockwise. For all Sort the drones, record the sorted drones, and reassign them with serial numbers to satisfy the following conditions: .

[0035] If a drone exists at the centroid, insert it sequentially between adjacent drones in the circular formation, generating an encirclement queue based on all possible insertion orders. Based on this, obtain all possible insertion orders I(J), and write all such candidate sequences into a FIFO candidate queue. middle.

[0036] If there are more than a set number of drones in the same direction, and the angle between densely packed drones in the same sector is less than a preset threshold, then the drone set is permuted and combined to obtain all possible insertion orders. These orders are then inserted into the interval positions of the initial encirclement order, and an encirclement queue is generated based on all possible insertion orders. Based on this, all possible insertion orders Π(J) are obtained. One order σ(J) is selected and inserted into the interval position of the original sequence to form a candidate adjacency sequence. All such candidate sequences are written into a FIFO candidate queue. middle.

[0037] In S53, the minimum capture cost corresponding to the capture queue is calculated. The specific expression is: In the formula, S For the encirclement and capture queue, For the initial angle offset, R The radius of the encirclement. The cost function; minimum capture cost The derivation steps are as follows: The desired position of each drone in the ideal circular encirclement formation is set as follows: In the formula, R The radius of the encirclement. The initial angle offset is used as an optimization parameter.

[0038] For candidate queues The encirclement cost is calculated for each adjacency case (i.e., a sorted sequence) in the array. The cost function is defined as: In the formula, For the encirclement queue S The Middle j The actual position of the drone is determined. Next, the parameters are solved for each candidate adjacency using the Gauss-Newton method. The minimum value of the function is iteratively solved using the second-order Taylor series expansion, the gradient, and the Hessian matrix. For the objective function f(x), the iterative formula of Newton's method is as follows: In the formula, For the first k Parameter estimates at the next iteration Given the Hessian matrix, the cost function to be minimized is for: Cost function To each Taking the partial derivative with R, we obtain the gradient as follows: The elements of the Hessian matrix are the second-order partial derivatives of the cost function, which can be expressed as: The parameters are updated using the Gauss-Newton method iterative formula, specifically expressed as follows: If the magnitude of the gradient is less than a given threshold, convergence is considered achieved, and iteration stops. The corresponding minimum containment cost is calculated as follows: S6 includes the following steps: S61. Obtain information about surrounding obstacles from the binocular camera, construct a three-dimensional grid or octree environment model, and map the continuous state of the UAV to discrete grid points and discrete heading angles for searching on the grid map. The continuous state includes position, orientation and speed. S62. Using the current position, orientation, and speed of the UAV as the starting point for the search, determine the safe encirclement point and attitude of the target area or target group according to the predetermined "capture and encircle" strategy. S63. In the discretized state space, the kinematic constraints of the UAV are used as the extended rules. A heuristic function is used to evaluate the cost from the current grid point to the target, which drives the search towards the target and generates a connected trajectory that satisfies the dynamic constraints. S64. Perform spline fitting or smoothing processing to minimize bending energy on the discrete trajectory generated by the hybrid state A algorithm to eliminate sharp turns in the broken line segment. Based on the speed and acceleration limits of the UAV, assign speed and timestamp to the smooth trajectory to obtain a continuous and executable time-parameterized trajectory. S65. Discretize the smooth trajectory and use the nearest obstacle distance detection algorithm to detect obstacle collisions at each sampling point. In response to the detection of a collision or insufficient safety distance, trigger local replanning and perform high and low altitude operations near that segment to correct the path in the smooth trajectory. S66. Integrate and splice the smooth trajectory, the time-parameterized trajectory, and the trajectory segments that have passed collision detection to generate the final capture trajectory sequence; S64 specifically refers to: Goal-oriented kinematic search front-end: Constructs a search tree based on the hybrid state A algorithm, generates motion primitives through discretized control input, and introduces a cost function during the search process. Using the target trajectory B(t) as heuristic information, a dual heuristic function is designed: In the formula, For dynamic distance estimation based on the optimal boundary value problem, For the drone target status, This is the current status of the drone. As weight, S t For the sum of the expected extended time, For path unfolding time, The following is obtained by extrapolating the predicted trajectory: In the formula, The Bézier curve trajectory of the original historical trajectory. The trajectory is obtained by extrapolation over time; this design enables the search process to have foresight regarding the target's movement, significantly improving search efficiency in complex environments.

[0039] Spatiotemporal trajectory optimization backend: Based on the initial path generated by the goal-oriented kinematic search frontend, a flight corridor composed of connected free-space cubes is constructed. ; In the formula, These are the control points of the Bézier curve. M The number of trajectory segments in the flight corridor; Establish an optimization problem with dynamic constraints under corridor constraints: In the formula, For smoothing terms, For the field of power, For dynamic constraint terms, q w intermediate waypoints T For the time of segmented trajectory, for x , y and z The combination x For segmented waypoints x Axis coordinates y For segmented waypoints y Axis coordinates z For segmented waypoints z Axis coordinates; This optimization problem is solved efficiently using the split convex optimization method, generating a time-parameterized trajectory; In this embodiment, the smoothing term is minimized using the third derivative integral, the potential field term uses a logarithmic barrier function to ensure trajectory safety, and the dynamic constraint term uses a piecewise linear penalty function to handle velocity and acceleration constraints. This optimization problem is efficiently solved using the split convex optimization method, generating a dynamically feasible trajectory that satisfies the requirements.

[0040] In S66, during high-speed encirclement operations, to reduce computational load and improve real-time performance, collision detection is performed on the next brief moment in the predicted path of the UAV. The specific collision detection method is as follows: For each UAV's time-parameterized trajectory, let the UAV... The trajectory is m Composition of segmental polynomials, defining the parameter set ; In the formula, For the first k Duration of segment , For the real number field, For the first k Segment polynomial order , For the matrix field, The coefficient matrix, , For the first k segment polynomial x Parameters in the axial direction, For the first k segment polynomial y Parameters in the axial direction, For the first k segment polynomial z Parameters in the axial direction; Calculate the relative time of the current moment within each segment using a polynomial. ; In the formula, It is the difference between the current time and the starting time. For absolute time, This is the starting timestamp of the trajectory. It is a function with maximum value. s The number of trajectory segments, For the minimum value index function, The time for each trajectory segment; Calculate the three-dimensional spatial position of the predicted trajectory ; In the formula, This is the current location of the drone. For the first s Duan Di j Term coefficient, , for x Coefficient in the axial direction, for y Coefficient in the axial direction, for z Coefficient in the axial direction, In the first s Relative time within a segment; In this embodiment, the present invention compares the sampling point of the local drone at each moment with the extrapolated position of all other drones at the same predicted moment. If the distance is lower than the preset safe distance, it is determined that there is a potential collision risk.

[0041] drones The sampling point at each moment in the predicted trajectory is compared with the position of the predicted trajectory of other drones at the same moment. If the distance is lower than the preset safe distance, it is determined that there is a potential collision risk. When a potential collision risk is identified, each drone with this risk receives the target's status information in real time and simultaneously acquires the position information of other drones. Each drone calculates its own Euclidean distance to the target and compares it with the Euclidean distances of other drones to the target. Drones are then sorted in ascending order based on their Euclidean distances to determine their movement priority. The drone with the highest priority maintains its planned trajectory, while the remaining drones perform obstacle avoidance maneuvers at different altitudes according to their priority ranking. The specific expression is: In the formula, For the first j The location coordinates of the drone Here are the target location coordinates, and rank(•) is the sorting function. In the description of this invention, it should be understood that the terms "center," "thickness," "upper," "lower," "horizontal," "top," "bottom," "inner," "outer," and "radial," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying the relative importance or the number of technical features implicitly specified. Therefore, a feature defined by "first," "second," and "third" may explicitly or implicitly include one or more of that feature.

Claims

1. A method for cooperative target capture by a swarm of unmanned aerial vehicles (UAVs) based on neuromorphic computing, characterized in that, The following steps are involved: S1. Based on the multi-view images obtained from the camera perspective of the UAV, the local pose and covariance of each UAV are obtained by using panoramic images, visual motion and grid-landmark cell models. S2. Based on the local pose and covariance of each UAV, the globally consistent relative pose of the UAV swarm is obtained through compressed broadcasting, common-view matching and graph optimization algorithms. S3. Based on the globally consistent relative pose of the UAV swarm, each UAV identifies the target, uses the unscented Kalman filter algorithm to estimate the target state, and fuses all the target state estimates to obtain the target state information in a unified coordinate system. S4. Based on historical target state information, use the Bézier curve generation algorithm to obtain a target motion trajectory under a historical state, expand the Bézier curve trajectory to predict the target motion, and obtain the target prediction state under a unified coordinate system. S5. Expand the drone swarm by extending the centroid to obtain the target capture queue. Use the Gauss-Newton method to solve for the minimum capture cost, and then obtain the optimal capture queue to generate the capture position of each drone. S6. Based on the capture position of each drone, a hybrid A* search algorithm is used to generate the drone's motion trajectory. Collision avoidance detection is performed based on the motion trajectory to generate the final capture trajectory sequence, thus completing the capture of the agile target.

2. The method for collaborative target encirclement and capture of unmanned aerial vehicles (UAVs) based on neuromorphic computing according to claim 1, characterized in that, S1 includes the following steps: S11. Based on the multi-view images obtained by the UAV cluster through the monocular camera, a continuous panoramic image is generated using the back projection and spherical mapping algorithm. S12. Based on the panoramic image, the ORB operator and RANSAC five-point method are used to obtain the rotation matrix and translation vector between adjacent frames, which are used as the measurement of visual motion. S13. Based on the feature description amplitude, the first pulse coding and grid-landmark cell model are used to perform position coding and memory; S14. Based on visual motion, IMU data and position encoding, UKF is used to predict and correct the state, and output the local pose and covariance of each UAV.

3. The method for cooperative target encirclement and capture of unmanned aerial vehicles (UAVs) based on neuromorphic computing according to claim 2, characterized in that, In S11, the method using back projection and spherical mapping algorithm is as follows: the multi-view image is back-projected onto the body coordinate system, wherein, for the ... i drone camera pixels The specific expression for back projection is: In the formula, These are the pixel coordinates projected onto the UAV's body coordinate system. Let the rotation matrix of the UAV in the world coordinate system be denoted as . The translation matrix of the starting point of the UAV in the world coordinate system. For the first i The intrinsic parameter matrix of the drone camera. Y i The coordinates are in the body coordinate system. u Let these be the coordinates of the image in the horizontal direction. v The coordinates of the image in the vertical direction; All Y i In radius Mapped onto the angular space on the sphere Panoramic images were synthesized using bilinear interpolation. ; S12 specifically involves: extracting key points and their descriptors from the panoramic image using the ORB operator, pairing them using FLANN, eliminating outliers using RANSAC, and then using the five-point method to calculate the relative motion under known intrinsic parameters to obtain the measured visual motion. ; In the formula, For rotation matrix, It is a translation vector; In S13, the grid-landmark cell model includes grid cells and landmark cell networks. Specifically, S13 maps feature description amplitudes to pulse times. , generating sparse pulses; In the formula, To prevent small amounts from being divided by zero, This is the upper limit of the encoding time window. For descriptors, The descriptor for the maximum amplitude at the current moment; When sparse impulses are input into the grid cell and landmark cell network, the grid cell state vector is updated using the following formula; In the formula, It is a non-linear activation. and For learnable parameters, For time t The grid cell state vector, For time t +1 grid cell state vector; Landmark cells activate associated memories by activating sparse pulse and grid cell state vectors according to the Hebbian rule; In the formula, To activate associated memories, To activate memories prior to association, For time t sparse pulses, For learning rate, It is the transpose symbol; In subsequent frames, correction parameters are established based on the memory associated with activation. Correct the grid code, and use the corrected grid code as the position code; In the formula, For learning rate, For time t +1 sparse pulse; In S14, the state vector of UKF and observation for: In the formula, For time t The local pose, For time t speed; The first estimate was obtained through UKF nonlinear prediction. Compared with the second forecast The Kalman gain is calculated based on the second prediction. ; In the formula, H The state transition equation is... This is the noise matrix; The UKF state vector is updated based on the Kalman gain and the first prediction, and the local pose and covariance of each UAV are output through the UKF. In the formula, This is the UKF-updated state vector. This is a transformation of the state transition equation.

4. The method for cooperative target encirclement and capture of unmanned aerial vehicles (UAVs) based on neuromorphic computing according to claim 3, characterized in that, S2 includes the following steps: S21. Based on covariance and sparse impulses, differential quantization or autoencoder algorithms are used to generate broadcast data packets for sharing neighboring machine observations. S22. Estimate the overlapping area of ​​the field of view based on the broadcast data packets, and use the ORB+RANSAC algorithm to calculate the relative observations between UAVs; S23. Based on the relative observations between UAVs, the incremental Gauss-Newton algorithm is used to solve the global pose and covariance of the UAVs. S24. Send the global pose and covariance of the UAV to each UAV, perform closed-loop correction of the global pose, and obtain the globally consistent relative pose of the UAV cluster.

5. The method for cooperative target encirclement and capture of unmanned aerial vehicle swarms based on neuromorphic computing according to claim 4, characterized in that, S21 specifically involves: performing differential quantization on the output local pose increment and covariance of each UAV to obtain the differentially quantized parameters. and : In the formula, A unified quantization function for vectors and matrices. For time t covariance, For time t +1 covariance, For local pose increments; In the formula, For time t -1 local pose; A sparse autoencoder is used to reduce the dimensionality of the grid cell state vector and sparse impulses to obtain the dimensionality reduction result. h ; In the formula, To concatenate vectors, For activation function, For sparse autoencoder parameters, State vector The corresponding estimated value; Establish broadcast data packets Broadcast data packets within the maximum latency allowed by link quality Reaching adjacent drones within the range; S22 specifically involves: obtaining the neighboring machine's grid activation by receiving broadcast data packets from the neighboring machine; calculating the field of view center deviation based on the UAV's own grid activation and the neighboring machine's grid activation; estimating the center and radius of the overlapping field of view circle; and obtaining the overlapping field of view area. Among them, the field of view center deviation includes the first spherical coordinate deviation. Second spherical coordinate deviation ; In the formula, Activate the drone's own mesh. Activate the neighboring machine's mesh. and This is a mapping function from grid encoding to spherical coordinate differences; ORB operators are used to extract segments of the decoded panoramic image in the overlapping regions of the field of view, and joint descriptors are constructed by combining the temporal information of the first pulse encoding. Matching pairs are obtained after RANSAC filtering. In the known camera intrinsic parameters K Solving the relative observations between UAVs under certain conditions This includes the rotation matrix of the drone itself relative to its neighboring drones. Translation vector ; In the formula, R For the first i The drone and the first j The difference in rotational variation between the drones For the first i A drone in the camera's internal reference K In this case, the difference in translation from the initial position, For the first j A drone in the camera's internal reference K In this case, the difference in translation from the initial position; Observation covariance of relative observations between UAVs The specific expression is: In the formula, The number of all drones to establish contact; S23 specifically involves: establishing a graph optimization model for a drone swarm, where the node set represents the local pose of each drone, the edge set consists of the relative observations and observation covariance among the drones, and the objective function of the graph optimization model is... The specific expression is: In the formula, For the first in three-dimensional space j The difference in pose between the drone and its initial position. For the first in three-dimensional space i The difference in pose between the drone and its initial position. for and The pose difference between them For SE(3) complex; The incremental Gauss-Newton algorithm works as follows: upon receiving new relative observations, it linearizes the residuals. The sparse information matrix is ​​incrementally updated, and the local pose of the UAV is updated after the increment is solved, which serves as the global pose of the UAV. In the formula, For the objective function The first partial derivative, For the objective function The first differential, For the objective function The second partial derivative, For the objective function The second differential, This is the difference between the sum of the partial derivative and the differential product; In the formula, The sparse information matrix before the update. This is the updated sparse information matrix; In the formula, p i This is the local pose of the drone before the update. This represents the updated local pose of the drone. S24 specifically involves broadcasting the global pose and covariance of the UAV to each UAV, and updating the UKF state of the local UAV as follows: In the formula, The updated state vector, This represents the global pose of the drone. This is the updated grid cell state vector. P i For covariance, and For the prior noise of velocity and mesh activation, To update the pose of the drone in three-dimensional space. For the updated speed; Various drones utilize and Restart the filtering loop and input the updated grid cell state vector and sparse impulses into the landmark cells for long-term memory updates to obtain the globally consistent relative pose of the UAV cluster.

6. The method for cooperative target encirclement and capture of unmanned aerial vehicle swarms based on neuromorphic computing according to claim 5, characterized in that, S3 includes the following steps: S31. Using the YOLO target recognition method, identify the target in the image information and obtain the target position in its coordinate system based on the image information; S32. Transform the target position using a rotation matrix to obtain the target position in a unified coordinate system; S33. Based on the target position in a unified coordinate system, UKF is used to estimate the target state, and the target state estimate and covariance after filtering out sensor errors are obtained. S34. Using a loosely coupled approach, the noise covariance of each UAV's observation of the target is combined to fuse all target state estimates and obtain target state information. S33 specifically refers to: Based on the state vector using UKF Covariance Matrix Generate 2 n The formula for generating +1 sigma point is as follows: In the formula, For sigma points, n Let be the dimension of the state vector. for K The estimated value corresponding to the position at a given time. For adjustment functions; In the formula, and To adjust the parameters; The predicted state mean is calculated by propagating the sigma point through a nonlinear state transition function. Covariance ; In the formula, To pass K The state value at time t is predicted. K The value at time +1, Q To measure the covariance matrix of the noise, , For the state part, For the noise part, and These are the weighting coefficients; In the formula, It is a tiny error; Map the predicted Sigma points to the measurement space and calculate the mean of the measurements. Covariance matrix S and cross-covariance matrix T ; In the formula, To pass K The observation values ​​at time 10 are predicted. K The value at time +1, R i This is the assumed pre-defined noise covariance matrix; Calculate the Kalman gain based on the covariance matrix. ; Based on the observed values and the mean of the measurements Update the target state estimate and covariance; updated target state estimate Covariance The expression is: In the formula, for K The observed value at time; In S34, the target state information is obtained. The specific expression is: In the formula, For the first i Target state estimation of the drone For the first The weight of the drone; In the formula, For the first The observation noise covariance of the drone.

7. The method for cooperative target encirclement and capture of unmanned aerial vehicle swarms based on neuromorphic computing according to claim 6, characterized in that, S4 includes the following steps: S41. Use the sliding window method to record historical target status information of a set length; S42. Based on historical target state information, generate a target motion trajectory under historical state according to the Bézier curve generation algorithm; S43. Calculate the predicted step size using the hyperbolic tangent function based on the distance between the centroid of the UAV cluster and the current target position. S44. Extend the predicted step size of the Bézier curve to obtain the predicted state of the target under a unified coordinate system. S42 specifically refers to: using Bézier curves to describe the target's motion trajectory, the target's motion trajectory... The specific expression is: In the formula, Given an nth-degree Bernstein polynomial basis, These are the control points for the Bézier curve; In S43, the prediction step size is calculated. The specific expression is: In the formula, For the pre-set prediction time step, d The distance to the target's current location. This is the current time.

8. The method for cooperative target encirclement and capture of unmanned aerial vehicles (UAVs) based on neuromorphic computing according to claim 7, characterized in that, S5 includes the following steps: S51. Calculate the centroid of the UAV swarm based on the globally unified coordinate system. S52. Calculate the vector from the centroid to the drone in each drone. All drones expand outward to form a capture queue. Sort the drones in the centroid region and the small fan-shaped region by permutation and combination to form several capture queues. S53. Use the Gauss-Newton method to solve the capture function and obtain the minimum capture cost corresponding to the capture queue. S54. Compare the minimum capture cost corresponding to all candidate capture queues, take the capture queue with the minimum minimum capture cost as the optimal capture queue, calculate the optimal capture point, and thus obtain the capture position of each drone. S52 specifically involves: obtaining the vector of each drone relative to its centroid, calculating the angle between this vector and the drone's encirclement direction vector, where the calculation of the... i The angle of the drone The specific expression is: In the formula, It is a modulo function. For vectors and x The included angle between the axes, with values ​​ranging from [0, 2π). For the first i The vector of the drone relative to its center of mass. , C Center of mass, , N For the number of drones, The direction vector for encirclement and capture. , T The target encirclement point; Sort all the drones by their included angles, assign serial numbers to the drones based on the sorting results, and generate a preliminary encirclement order. If there is a drone at the centroid, insert the drone into the circular formation between two adjacent drones in sequence, and generate an encirclement queue according to all possible insertion orders. If there are more than a set number of drones in the same direction, and the angle between the dense drones in the same sector is less than a preset threshold, then the drone set is arranged and combined to obtain all possible insertion orders, and inserted into the interval position of the initial encirclement order. An encirclement queue is generated according to all possible insertion order. In S53, the minimum capture cost corresponding to the capture queue is calculated. The specific expression is: In the formula, S For the encirclement and capture queue, For the initial angle offset, R The radius of the encirclement. The cost function; In the formula, For the encirclement queue S The Middle j The actual location of the drone.

9. The method for cooperative target encirclement and capture of unmanned aerial vehicles (UAVs) based on neuromorphic computing according to claim 8, characterized in that, S6 includes the following steps: S61. Obtain information about surrounding obstacles from the binocular camera, construct a three-dimensional grid or octree environment model, and map the continuous state of the UAV to discrete grid points and discrete heading angles for searching on the grid map. The continuous state includes position, orientation and speed. S62. Using the current position, orientation, and speed of the UAV as the starting point for the search, determine the safe encirclement point and attitude of the target area or target group according to the predetermined "capture and encircle" strategy. S63. In the discretized state space, the kinematic constraints of the UAV are used as the extended rules. A heuristic function is used to evaluate the cost from the current grid point to the target, which drives the search towards the target and generates a connected trajectory that satisfies the dynamic constraints. S64. Perform spline fitting or smoothing processing to minimize bending energy on the discrete trajectory generated by the hybrid state A algorithm to eliminate sharp turns in the broken line segment. Based on the speed and acceleration limits of the UAV, assign speed and timestamp to the smooth trajectory to obtain a continuous and executable time-parameterized trajectory. S65. Discretize the smooth trajectory and use the nearest obstacle distance detection algorithm to detect obstacle collisions at each sampling point. In response to the detection of a collision or insufficient safety distance, trigger local replanning and perform high and low altitude operations near that segment to correct the path in the smooth trajectory. S66. Integrate and splice the smooth trajectory, the time-parameterized trajectory, and the trajectory segments that have passed collision detection to generate the final capture trajectory sequence; S64 specifically refers to: Goal-oriented kinematic search front-end: Constructs a search tree based on the hybrid state A algorithm, generates motion primitives through discretized control input, and introduces a cost function during the search process. Using the target trajectory B(t) as heuristic information, a dual heuristic function is designed: In the formula, For dynamic distance estimation based on the optimal boundary value problem, For the drone target status, This is the current status of the drone. As weight, S t For the sum of the expected extended time, For path unfolding time, The following is obtained by extrapolating the predicted trajectory: In the formula, The Bézier curve trajectory of the original historical trajectory. The trajectory obtained by extrapolating time; Spatiotemporal trajectory optimization backend: Based on the initial path generated by the goal-oriented kinematic search frontend, a flight corridor composed of connected free-space cubes is constructed. ; In the formula, These are the control points of the Bézier curve. M The number of trajectory segments in the flight corridor; Establish an optimization problem with dynamic constraints under corridor constraints: In the formula, For smoothing terms, For the field of power, For dynamic constraint terms, q w intermediate waypoints T For the time of segmented trajectory, for x , y and z The combination x For segmented waypoints x Axis coordinates y For segmented waypoints y Axis coordinates z For segmented waypoints z Axis coordinates; This optimization problem is solved efficiently using the split convex optimization method, generating a time-parameterized trajectory; In S66, the collision detection method is as follows: For each UAV's time-parameterized trajectory, let the UAV... The trajectory is m Composition of segmental polynomials, defining the parameter set ; In the formula, For the first k Duration of segment , For the real number field, For the first k Segment polynomial order , For the matrix field, The coefficient matrix, , For the first k segment polynomial x Parameters in the axial direction, For the first k segment polynomial y Parameters in the axial direction, For the first k segment polynomial z Parameters in the axial direction; Calculate the relative time of the current moment within each segment using a polynomial. ; In the formula, It is the difference between the current time and the starting time. For absolute time, This is the starting timestamp of the trajectory. It is a function with maximum value. s The number of trajectory segments, For the minimum value index function, The time for each trajectory segment; Calculate the three-dimensional spatial position of the predicted trajectory ; In the formula, This is the current location of the drone. For the first s Duan Di j Term coefficient, , for x Coefficient in the axial direction, for y Coefficient in the axial direction, for z Coefficient in the axial direction, In the first s Relative time within a segment; drones The sampling point at each moment in the predicted trajectory is compared with the position of the predicted trajectory of other drones at the same moment. If the distance is lower than the preset safe distance, it is determined that there is a potential collision risk. When a potential collision risk is identified, each drone with this risk receives the target's status information in real time and simultaneously acquires the position information of other drones. Each drone calculates its own Euclidean distance from the target and compares it with the Euclidean distances of other drones to the target. Drones are then sorted in ascending order based on their Euclidean distances to determine their movement priority. The drone with the highest priority maintains its planned trajectory, while the remaining drones perform obstacle avoidance at different altitudes according to their priority ranking. The specific expression is: In the formula, For the first j The location coordinates of the drone Here are the target location coordinates, and rank(•) is the sorting function.

Images