Unmanned aerial vehicle autonomous obstacle avoidance and countermeasure method and device based on multi-sensor fusion

Abstract

The application belongs to the field of unmanned aerial vehicle control, and relates to an unmanned aerial vehicle autonomous obstacle avoidance and countermeasure method and device based on multi-sensor fusion, which comprises the following steps: synchronizing laser radar, millimeter wave radar, binocular vision and ultrasonic sensor in time and space, dynamically evaluating confidence and adopting adaptive weighted fusion to construct a three-dimensional map; identifying and tracking dynamic obstacles, predicting trajectories and calculating collision probability to realize local obstacle avoidance and re-planning; passively monitoring communication frequency bands, extracting radio frequency fingerprints and comparing with a white list to identify suspicious unmanned aerial vehicle identity and direction; forming a directional beam to emit tracking interference or decoy navigation signals; online optimizing fusion weights and path planning parameters on an edge computing unit; and multi-dimensionally evaluating countermeasure effects, automatically adjusting parameters if the effects are below a threshold value and storing cases in a knowledge base for strategy migration. The application improves perception reliability and environmental adaptability, and has high intelligentization and precision of obstacle avoidance and countermeasures, and has the ability of continuous evolution and closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and apparatus for autonomous obstacle avoidance and countermeasures by UAVs based on multi-sensor fusion. Background Technology

[0002] Existing autonomous obstacle avoidance and countermeasure methods for drones suffer from problems such as isolated sensing, passive response, and insufficient intelligence. The specific shortcomings of existing technologies are reflected in the following four aspects: First, multi-sensor fusion mechanisms are rigid and lack environmental adaptability. Most existing technologies employ sensor fusion strategies with fixed weights or simple switching. For example, visual sensors are prone to failure in strong light or rain / fog environments, while LiDAR accuracy decreases in smoke. However, fixed-weight fusion algorithms cannot dynamically adjust their confidence levels, leading to significant errors or gaps in 3D map construction. This demonstrates the shortcomings of existing technologies in terms of robustness to dynamic environmental perception.

[0003] Second, obstacle avoidance planning lacks the ability to predict dynamic targets. Traditional obstacle avoidance algorithms mostly rely on the current obstacle position for reactive avoidance, failing to adequately identify and track the short-term trajectories of dynamic obstacles (such as birds and other drones). When facing high-speed maneuvering targets, the lag in calculating collision probabilities often leads to unreasonable obstacle avoidance paths or obstacle avoidance failures. This reflects its shortcomings in predicting dynamic obstacle behavior and proactive obstacle avoidance.

[0004] Third, countermeasures are limited and lack closed-loop feedback. Existing countermeasures often employ omnidirectional electromagnetic interference or rely solely on radio frequency detection, which can easily cause electromagnetic pollution and fails to accurately distinguish between friendly and enemy drones. Furthermore, the actual effectiveness of jamming signals is often not assessed after transmission, resulting in countermeasure parameters (such as power and direction) not being able to adaptively adjust. This can lead to situations where jamming is ineffective but transmission continues or friendly fire occurs. This exposes the shortcomings of existing technologies in terms of precise and intelligent countermeasures and self-evaluation of effectiveness.

[0005] Fourth, it lacks online learning and knowledge transfer capabilities. Most drone systems have their perception and decision-making parameters fixed after leaving the factory, making it impossible to optimize and fuse weights or path planning parameters online based on successful or failed obstacle avoidance / countermeasure cases accumulated during actual flight. For different scenarios (such as urban canyons and forests), the system requires manual re-tuning and lacks the ability to quickly transfer strategies from historical experience. This demonstrates its significant shortcomings in edge-end online adaptation and scenario transfer learning. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles (UAVs) based on multi-sensor fusion, employing the following technical solution, including the following steps: The system performs time and space synchronization of airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors, dynamically evaluates the confidence of each sensor according to environmental changes, and uses an adaptive weighted fusion algorithm to build a three-dimensional environment map in real time. Based on the three-dimensional environment map, dynamic obstacles are identified and tracked, their short-term motion trajectories are predicted, collision probability is calculated, and local obstacle avoidance paths are replanned when there is a risk of collision. Passively monitor the communication frequency band of drones, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of suspicious drones, and estimate their direction of arrival and distance. Based on the identified location and signal characteristics of the suspicious drone, a directional beam is formed pointing to the suspicious drone, a tracking jamming signal is emitted to cut off its communication link, and / or a deceptive navigation signal is sent to guide it off course. On the airborne edge computing unit, based on historical flight data and real-time obstacle avoidance and countermeasure effects, the fusion weights and path planning parameters in the adaptive weighted fusion algorithm are optimized online. The countermeasures are evaluated quantitatively from multiple dimensions. If the evaluation results are lower than the preset threshold, the countermeasure parameters are automatically adjusted, and successful or unsuccessful cases are stored in the knowledge base for strategy migration in similar scenarios in the future.

[0007] To address the aforementioned technical problems, this invention also provides an autonomous obstacle avoidance and countermeasure device for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, employing the following technical solution, including: The fusion mapping module is used to synchronize airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors in time and space, and dynamically evaluate the confidence of each sensor according to environmental changes. It uses an adaptive weighted fusion algorithm to build a three-dimensional environment map in real time. The path replanning module is used to identify and track dynamic obstacles based on the three-dimensional environment map, predict their short-term motion trajectory, calculate the collision probability, and replan the local obstacle avoidance path when there is a risk of collision. The identity identification module is used to passively monitor the communication frequency band of drones, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of suspicious drones, and at the same time estimate their incoming wave direction and distance. The collaborative countermeasure module is used to form a directional beam pointing at the suspicious drone based on the identified location and signal characteristics of the suspicious drone, transmit tracking jamming signals to cut off its communication link, and / or send deceptive navigation signals to guide it off course. The online optimization module is used to optimize the fusion weights and path planning parameters in the adaptive weighted fusion algorithm online on the airborne edge computing unit based on historical flight data and real-time obstacle avoidance and countermeasure effects. The self-evolution module is used to perform multi-dimensional quantitative evaluation of countermeasures. If the evaluation result is lower than the preset threshold, the countermeasure parameters will be automatically adjusted, and successful or unsuccessful cases will be stored in the knowledge base for strategy migration in similar scenarios in the future.

[0008] To address the aforementioned technical problems, the present invention also provides a computer device that employs the technical solution described below, comprising a memory and a processor. The memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the aforementioned method for autonomous obstacle avoidance and countermeasures of unmanned aerial vehicles based on multi-sensor fusion.

[0009] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium, which employs the technical solution described below. The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the aforementioned method for autonomous obstacle avoidance and countermeasures of unmanned aerial vehicles based on multi-sensor fusion.

[0010] Compared with the prior art, the present invention has the following main advantages: (1) Significantly improves perception reliability and environmental adaptability: By dynamically evaluating the confidence of each sensor and adopting adaptive weighted fusion, the system can automatically reduce the weight of low-quality sensors in complex or sudden environments (such as rain, fog, strong backlight, electromagnetic interference), ensuring the continuity and accuracy of the three-dimensional environment map, and greatly enhancing the survivability of UAVs in unstructured environments.

[0011] (2) High level of intelligence and precision in obstacle avoidance and countermeasures: It can predict the short-term trajectory of dynamic obstacles and calculate the collision probability, so as to achieve proactive obstacle avoidance rather than blind abrupt turns, thereby improving flight safety and path smoothness. At the same time, through directional beam tracking jamming, it can accurately distinguish between friend and foe, and carry out surgical countermeasures against suspicious UAVs, avoiding broad-spectrum energy leakage and impact on surrounding equipment.

[0012] (3) Possesses continuous evolution and closed-loop optimization capabilities: Online optimization of historical parameters is achieved on the airborne edge computing unit, and countermeasures can be automatically adjusted based on multi-dimensional quantitative evaluation results, forming a closed loop of perception-decision-action-evaluation-optimization. After successful and failed cases are stored in the knowledge base, strategies for similar scenarios can be quickly migrated, making the UAV system more and more intelligent with use, and significantly reducing the cost of manual operation and maintenance and parameter debugging. Attached Figure Description

[0013] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1 This is a flowchart of an embodiment of the UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion of the present invention; Figure 2 This is a schematic diagram of an embodiment of the UAV autonomous obstacle avoidance and countermeasure device based on multi-sensor fusion of the present invention. Figure 3 This is a schematic diagram of the structure of an embodiment of the computer device of the present invention. Detailed Implementation

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects and not to describe a particular order.

[0016] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0018] It should be noted that the UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion provided in the embodiments of the present invention is generally executed by a server / terminal device, and correspondingly, the UAV autonomous obstacle avoidance and countermeasure device based on multi-sensor fusion is generally set in the server / terminal device.

[0019] It should be understood that the number of terminal devices, networks, and servers is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used. Example

[0020] Please refer to Figure 1 The flowchart illustrates an embodiment of the UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion according to the present invention. The UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion includes the following steps: Step S1: Time and space synchronization of airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors, and dynamic evaluation of the confidence of each sensor according to environmental changes, and real-time construction of a three-dimensional environment map using an adaptive weighted fusion algorithm.

[0021] In this embodiment, the electronic device (e.g., a server / terminal device) running on the UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion can receive the UAV autonomous obstacle avoidance and countermeasure request based on multi-sensor fusion via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods.

[0022] In this embodiment, step S1 may specifically include the following steps: S11 triggers the acquisition of data from each sensor based on the timing signal from the Global Navigation Satellite System, and performs cubic spline interpolation on the measurements of sensors with low sampling rates to achieve coarse time synchronization; it projects all sensor data onto a unified world coordinate system through a pre-calibrated rotation matrix and translation vector to achieve spatial synchronization.

[0023] The airborne GPS / BeiDou timing module outputs a 1PPS (Pulse Per Second) signal with a rising edge accuracy better than 50ns, simultaneously triggering hardware acquisition from lidar (50Hz), millimeter-wave radar (30Hz), binocular camera (30Hz), and ultrasonic sensor (20Hz).

[0024] For ultrasound with the lowest sampling rate, the time axis t of the highest frequency lidar is used. LiDAR Using [n] as the reference, at two adjacent ultrasound sampling points t j and t j+1 Cubic spline interpolation is performed between the points to obtain the virtual ultrasonic distance value at the corresponding time.

[0025] Regarding spatial synchronization: The rotation matrix R from the four sensor coordinate systems to the body coordinate system was obtained in advance using the Zhang Zhengyou calibration method. i Translation vector t iProject all point clouds onto the world coordinate system P with the UAV's center of gravity as the origin. world .

[0026] The cubic spline interpolation formula is: S(t) = a j (tt j ) 3 +b j (tt j ) 2 +c j (tt j )+d j , t∈[t j ,t j+1 ].

[0027] Where: S(t): The sensor measurement value at time t after interpolation (e.g., ultrasonic ranging value, in meters). j , t j+1 Two adjacent original sampling times (in seconds). j b j c j d j The cubic spline coefficients on the j-th interval are uniquely determined by the continuity of the function value, the continuity of the first derivative, the continuity of the second derivative, and the natural boundary conditions.

[0028] Ensure that the data after time synchronization is smooth and without abrupt changes, and avoid false obstacles appearing on the map due to time deviation.

[0029] The formula for spatial projection is: P world =R i P sensor,i +t i .

[0030] Where: P sensor,i : The coordinates of the point in the coordinate system of the i-th sensor (3×1 vector). R i : 3×3 rotation matrix, obtained from calibration. t i : 3×1 translation vector.

[0031] Unify all sensor data into the same coordinate system to facilitate subsequent fusion.

[0032] The purpose of step S11 is to eliminate the spatiotemporal misalignment caused by the different sampling frequencies of each sensor and the transmission delay.

[0033] S12 calculates the historical error variance and current signal-to-noise ratio of each sensor in real time, and calculates the adaptive fusion weight of each sensor based on the exponential weighting method, so that the measurement values ​​of each sensor are weighted and averaged to obtain the fused environmental perception data.

[0034] The confidence level w of each sensor i at time k i (k) is determined by its historical error variance σ i 2 (k) and the current signal-to-noise ratio (SNR) i (k) Joint decision.

[0035] The historical error variance is obtained by recursion using the Kalman filter residuals: σ i 2 (k)=(1-α)σ i 2 (k-1)+α(z i (k)-z ̂(k-1)) 2 , where α=0.1 is the forgetting factor.

[0036] Signal-to-noise ratio (SNR): Millimeter-wave radar is based on the ratio of echo amplitude to noise floor; binocular vision is based on the ratio of image patch contrast to average grayscale; lidar is based on the ratio of reflectivity to minimum detectable reflectivity; and ultrasound is based on the ratio of echo amplitude to threshold.

[0037] A fuzzy adaptive weighted fusion algorithm is adopted, and the weights are updated every 0.1 seconds. The fused distance estimate ẑ(k) is a weighted average.

[0038] The adaptive weighted fusion formula is: ; .

[0039] in: : The fused distance estimate (in meters). z i (k): The raw distance measurement (m) of the i-th sensor at time k. σ i 2 (k): The estimated variance (m²) of the i-th sensor, characterizing its historical stability. SNR i (k): Signal-to-noise ratio (SNR) of the i-th sensor (dimensionless), ranging from 0.1 to 10. β: Coefficient adjusting the weight between variance and SNR, set to 0.7 (experimental calibration value). exp(⋅): Exponential function that amplifies the weight of high SNR sensors.

[0040] When visual performance deteriorates due to poor lighting at night (SNR drops from 5 to 0.5), its weight automatically decreases from 0.4 to 0.05; millimeter-wave radar maintains stable performance in rainy weather, so its weight increases to achieve adaptive fault tolerance.

[0041] The function of step S12 is to adjust the fusion weights of each sensor in real time according to environmental changes (rain, fog, light, strong reflection).

[0042] S13 maps the fused point cloud data to an octree structure and uses a Bayesian filter in the form of log probability ratio to update the occupancy probability of each voxel, forming a three-dimensional occupancy probability map.

[0043] The OctoMap open-source framework is used, with a map resolution of 0.2 m × 0.2 m × 0.2 m and a maximum depth of 16 layers.

[0044] Each voxel stores an occupancy probability P(n|z). 1:k The binomial Bayesian filter is used to update the value.

[0045] LiDAR point clouds are projected by ray casting: voxels along the ray direction from the sensor origin are marked as free (probability decreases) and voxels encountered at the endpoint are marked as occupied (probability increases).

[0046] For millimeter-wave radar and ultrasound, only the corresponding single voxel is updated (because of the low angular resolution).

[0047] The formula for updating the log-probability ratio is: L(n|z 1:k )=L(n|z 1:k-1 )+L(n|z k ); P(n|z 1:k )=1 / (1+exp(-L(n|z 1:k )).

[0048] Where: L(n|z 1:k L(n|z) represents the log-odds ratio of voxels n, initially set to L0 = 0 (corresponding to a probability of 0.5). k ): Based on the logarithmic update obtained from the k-th measurement, if the lidar detects an obstacle, then L occ =+0.8, if the ray passes through the empty region then L free =-0.4. exp(⋅): Exponential function.

[0049] The probability update is converted into an addition operation, which is fast and suitable for edge computing (approximately 5μs per voxel for each update). In the final map, P>0.7 is considered an obstacle, P<0.2 is considered free space, and the area in between is the unknown region.

[0050] The purpose of step S13 is to transform the fused sparse point cloud into a memory-efficient octree map with queryable occupancy probabilities.

[0051] The purpose of step S1 is to resolve the inconsistencies in time and space among the four types of sensors: lidar, millimeter-wave radar, binocular vision, and ultrasonic sensors. It uses an adaptive weighted fusion algorithm to construct a high-precision 3D environment map in real time, providing a unified coordinate reference for subsequent obstacle avoidance and countermeasures.

[0052] Step S2: Identify and track dynamic obstacles based on the 3D environment map, predict their short-term motion trajectory, calculate the collision probability, and replan the local obstacle avoidance path when there is a risk of collision.

[0053] In this embodiment, step S2 may specifically include the following steps: S21 removes the UAV's own motion by point cloud registration, clusters the remaining motion vectors to obtain dynamic obstacles, and uses extended Kalman filtering to estimate and predict the position, velocity and acceleration of each dynamic obstacle.

[0054] The point clouds of two adjacent frames are registered using the ICP (Iterative Closest Point) algorithm. The motion of the UAV itself is removed, and the remaining motion vectors are clustered (Euclidean clustering) to obtain the dynamic obstacles.

[0055] For each dynamic obstacle, an Extended Kalman Filter (EKF) model is established, with the state vector being x=[x,y,z,v]. x ,v y ,v z ,a x ,a y ,a z ] T There are a total of 9 dimensions.

[0056] The process model is a uniform acceleration model (CA model), and the observations are provided by millimeter-wave radar (range, azimuth, radial velocity) and binocular vision (boundary box center, depth).

[0057] The EKF prediction formula is: x ̂ k|k-1 =F k x ̂ k-1|k-1 ; .

[0058] Where: x ̂ k-1|k-1 Predict the state (9×1 vector) at time k based on the estimate at time k-1. F k State transition matrix (9×9), Δt=0.05s (synchronized with lidar frequency). Q k Process noise covariance matrix (diagonal matrix, with position / velocity / acceleration noise set to 0.01, 0.1, 0.5).

[0059] The EKF update formula is: K k =P k|k-1 H k T (H k P k|k-1 H k T +R k ) -1 ; x ̂ k|k =x ̂ k|k-1 +K k (z k -H k x ̂ k|k-1 ).

[0060] Where: K k Kalman gain (a 9×m matrix, where m is the observation dimension). P k|k-1 Predicted covariance (9×9). H k : Observation matrix (m×9), for example, extracting the position and velocity correspondences from the state. R k : Observation noise covariance (m×m, calibrated in real time by the sensor). k : Actual observation vector (m×1).

[0061] It can predict the position of dynamic obstacles 0.5 seconds in advance, with a prediction error of < 0.1m.

[0062] The purpose of step S21 is to separate dynamic obstacles from the point cloud and estimate their position, velocity, and acceleration.

[0063] S22, calculate the relative position vector and relative velocity vector between the UAV and the dynamic obstacle, determine whether there is a collision risk based on the collision cone condition, and calculate the collision time. When the collision time is lower than the safety threshold, mark the obstacle as a dangerous obstacle.

[0064] The drone is considered to have a radius r u A sphere with a radius of 0.5m is considered as an obstacle. o The sphere (estimated based on clustered point cloud).

[0065] Calculate the relative position p rel =p o -p u and relative velocity v rel =v u -v o .

[0066] Define a collision cone: when ∠(p rel ,v rel ) <arcsin((r u +ro ) / |p rel There is a risk of collision when |).

[0067] Calculate the collision time (TTC).

[0068] The formula for calculating the time to collision (TTC) is: .

[0069] Where: p rel : Relative position vector (3×1), p o p represents the location of the obstacle. u This represents the drone's location, in meters (m). rel : Relative velocity vector (3×1), unit m / s. ⋅ : Dot product operation. |⋅| : Magnitude of the vector. r u =0.5m: UAV radius. o : Obstacle radius (calculated from point cloud clustering), in meters.

[0070] If the Time To Collision (TTC) is less than 2 seconds and the collision cone condition is met, the obstacle is marked as hazardous, triggering a replanning process. The TTC calculation formula can handle non-center-to-center collisions.

[0071] The purpose of step S22 is to assess the probability of collision between the current flight path and the predicted dynamic obstacles.

[0072] S23 uses a high-order Bézier curve to represent the local path, constructs a cost function with path length, curvature and safe distance from obstacles, and obtains a new path that satisfies the dynamic constraints of the UAV and avoids dangerous obstacles through optimization.

[0073] The new path is represented by a fifth-order Bézier curve, defined by six control points Q0, Q1, ..., Q5.

[0074] Constraints: Start point Q0 = current position, end point Q5 = local target point (5 meters ahead along the global path), and the first derivative (velocity direction) is continuous.

[0075] The objective function J is minimized, which includes path length, squared curvature integral, and distance penalty to obstacles.

[0076] The solution is obtained using Sequential Quadratic Programming (SQP), and convergence occurs within 50 iterations.

[0077] The formula for calculating Bézier curves is: , t∈[0,1].

[0078] Where: B(t): the coordinates of the point on the curve corresponding to parameter t (3×1 vector), t=0 is the starting point, and t=1 is the ending point. : Binomial coefficient. Q i : The i-th control point (3×1 vector).

[0079] Fifth-order Bézier curves ensure the continuity of position, velocity, and acceleration (C² continuity), making them suitable for UAV flight control execution.

[0080] The formula for optimizing the cost function is: .

[0081] Where: B'(t)=dB / dt: first derivative (velocity vector), unit m / s.

[0082] κ(t)=|B'(t)×B″(t)| / |B'(t)| 3 Curvature (unit: 1 / m), × represents the cross product.

[0083] d(B(t),O j d: The minimum Euclidean distance between the path point and the j-th obstacle (static or dynamically predicted position), in meters. safe =1.5m: Safety distance. λ1=0.3, λ2=0.4, λ3=0.3: Weighting coefficients.

[0084] The generated path satisfies all constraints, minimizing path length (energy saving), minimizing curvature (smooth flight), and penalizing safe distances for intruding into obstacles. Computation time < 20ms.

[0085] The purpose of step S23 is to regenerate a smooth, dynamically feasible, obstacle-avoiding local path after detecting a high collision probability.

[0086] The purpose of step S2 is to identify dynamic obstacles based on the 3D map obtained in step S1, predict their short-term movement trajectories, and replan the obstacle avoidance path.

[0087] Step S3: Passively monitor the communication frequency band of the drone, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of the suspicious drone, and at the same time estimate its incoming wave direction and distance.

[0088] In this embodiment, step S3 may specifically include the following steps: S31 performs a fast Fourier transform on the received broadband signal to obtain the power spectrum, uses constant false alarm rate detection to extract suspicious frequency points, and performs independent component analysis on the baseband signal at the suspicious frequency points to separate the signal components from different UAV sources from the mixed signal.

[0089] Using a LimeSDR or AD9361 chip, with a sampling rate of 20 Msps, perform a 2048-point FFT every 0.1 seconds to obtain the power spectral density.

[0090] Constant False Alarm Rate (CFAR) Detection: Sliding window estimation of noise basis, threshold P th =μ noise +5σ noise Frequency points exceeding the threshold are marked as suspicious.

[0091] Independent component analysis (ICA) was used to separate the baseband signals at the suspicious frequency points, assuming that the number of observation channels M=4 (4 antennas) and the number of source signals N=2~3.

[0092] The ICA separation formula is: x(t) = As(t) + n(t); y(t)=Wx(t).

[0093] Where: x(t): M×1 received mixed signal vector (M=4). s(t): N×1 source signal vector (communication signals of each UAV). A: M×N unknown mixing matrix. n(t): M×1 noise vector. W: N×M demixing matrix, achieved by maximizing the negative entropy J(y)∝[E{G(y)}-E{G(ν)}]. 2 Learned. Where: G is a non-quadratic function. y(t): the separated independent signal components (N×1).

[0094] In scenarios where multiple drones are flying simultaneously in a city, signals from different sources can be separated to avoid misjudgment.

[0095] The purpose of step S31 is to extract suspicious signal features from the complex electromagnetic background and separate the communication signals of multiple UAVs.

[0096] S32 extracts the frequency hopping period, modulation method, and demodulated data frame address from the separated signal components and compares them with the built-in drone RF fingerprint database and local whitelist. If they cannot match and the flight behavior is abnormal, the drone is identified as suspicious.

[0097] From the separated signal y i Extracting the frequency hopping period from (t): Perform a short-time Fourier transform (STFT) on the signal, statistically analyze the peak sequence of frequency changes over time, calculate the histogram of adjacent hopping time intervals, and take the modulus value as T̂. hop .

[0098] Extracting the modulation method: Calculating the higher-order cumulant C 20 C 21Distinguish between OFDM / QPSK / BPSK by comparing with theoretical values.

[0099] For non-frequency hopping signals (such as Wi-Fi video transmission), demodulate the data frame and extract the MAC address (the first 24 bits of the OUI correspond to the manufacturer).

[0100] If the MAC address is not on the whitelist and the flight altitude is greater than 50m or the MAC address enters a no-fly zone, it is considered "suspicious" by comparing it with the built-in database and the local whitelist.

[0101] The formula for estimating the frequency hopping period is: R(τ) = ∫x(t) x* (t+τ) e -j2πfcτ ) dt; T ̂ hop =arg max|R(τ)|,τ≠0.

[0102] Where: x(t): the separated complex baseband signal. x*(t+τ): the complex conjugate after a time delay of τ. f c : Center frequency estimate (obtained from the FFT peak). R(τ): Delay correlation function. T̂ hop : Estimated frequency hopping period (in seconds).

[0103] Many unauthorized drones use fixed-parameter frequency hopping, and the frequency hopping cycle can be identified for tracking interference in step 4.

[0104] The purpose of step S32 is to identify the type of drone and its MAC address from which the separated signal originates.

[0105] S33 uses an airborne array antenna to receive signals, estimates the direction of arrival of the suspicious UAV through a subspace decomposition algorithm, and estimates its distance based on the received signal strength and free space propagation model.

[0106] Array antenna: 4-element uniform circular array with radius r=λ / 2 (λ=12.5cm), and the direction of arrival (DOA) is estimated by the MUSIC algorithm.

[0107] Calculate the covariance matrix R of the received signal. xx =E[x(t) x H [(t)], perform eigenvalue decomposition, noise subspace E N Corresponding to small eigenvalues.

[0108] Scan azimuth θ∈[0,2π), elevation ϕ∈[-π / 2,π / 2], calculate spatial spectrum P MUSIC The peak value corresponds to the DOA.

[0109] Distance estimation: Measure the received signal strength RSSI (dBm) and use the free space propagation model to inversely estimate the distance.

[0110] The formula for calculating the MUSIC spatial spectrum is: P MUSIC (θ,ϕ)=1 / (a H (θ,ϕ) E N E N H a(θ,ϕ)).

[0111] Where: a(θ,ϕ): array steering vector (4×1), the m-th element is exp(j×(2π / λ)×rsinθcos(ϕ-γ)). m )), γ m E represents the azimuth angle of the array element. N : 4×(4-N) src The eigenvector matrix of the noise subspace, N src The number of source signals (usually 1 to 2). E N H : Conjugate transpose.

[0112] The smaller the denominator, the higher the spectral peak. The peak position (θ̂, ϕ̂) indicates the direction of arrival. The angular resolution is approximately 3°.

[0113] The distance estimation formula for the free-space propagation model is: .

[0114] Where: R: distance (km). P t Suspicious drone transmit power (assumed to be 20 dBm, typical value). r Airborne received power (dBm, measured by RSSI). G t =2dBi: Suspicious drone antenna gain. G r =3dBi: Airborne receiving antenna gain. f: Center frequency (MHz, e.g., 2450). Constant 32.44 from 20log 10 Unit conversion for (4π / c).

[0115] Within 1.5 km, the distance error is less than 20%, which can support power control of directional interference.

[0116] The purpose of step S33 is to estimate the distance and direction of the suspicious drone using only an airborne single station.

[0117] Step S4: Based on the identified location and signal characteristics of the suspicious drone, a directional beam is formed pointing to the suspicious drone, a tracking jamming signal is emitted to cut off its communication link, and / or a deceptive navigation signal is sent to guide it off course.

[0118] In this embodiment, step S4 may specifically include the following steps: S41, based on the estimated direction of arrival, calculate the phase offset of each transmission channel of the phased array antenna, so that the main lobe of the synthesized beam is pointed at the suspicious UAV, and adjust the transmission power according to the estimated distance so that the strength of the interference signal received at the target reaches the preset threshold.

[0119] 4. Transmit channel phased array antenna, operating frequency 2.4GHz or 5.8GHz, element spacing d=λ / 2.

[0120] Based on the DOA angle (θ̂, ϕ̂) estimated in step S3, calculate the phase offset Δφ for each channel. m This aligns the main lobe of the beam with the target.

[0121] Adjust the transmission power P according to the distance R. tx Make the interference signal strength at the target J=P tx +G t +G r -20log 10 (R)-20log 10 (f)-32.44≥-50dBm.

[0122] The formula for calculating the phase of a phased array is: Δφ m =2π⋅d m ⋅u(θ,ϕ) / λ.

[0123] Where: λ=c / f: wavelength (c=3×10) 8 m / s). ⋅: Dot product operation.

[0124] d m : The position vector of the m-th antenna element relative to the reference point (3×1), in units of m.

[0125] u(θ,ϕ)=[sinθcosϕ,sinθsinϕ,cosθ] T : Target direction unit vector.

[0126] By adjusting the phase, signals from different transmitting antennas are superimposed in phase at the target, achieving a directional gain of 15 dBi with sidelobes below -15 dB.

[0127] The formula for adjusting the transmit power is: P tx =min(P max J req -(G t +G r -20log 10 R-20log10 f-32.44)).

[0128] Where: P tx Transmit power (dBm). P max =30dBm (1W, legal limit). J req =-50dBm: The required interference intensity at the target location.

[0129] This ensures both the effectiveness of the jamming and avoids unnecessary high-power transmissions.

[0130] The purpose of step S41 is to concentrate the interference energy towards the suspected drone, so as to avoid interference with other devices.

[0131] S42 tracks the frequency hopping frequency or fixed operating frequency of suspicious drones in real time, and transmits a narrowband jamming signal with the same frequency and superior power as the target signal during each dwell time, causing the bit error rate of its communication link to rise above the interruption threshold.

[0132] Step S3: Real-time output of the instantaneous frequency f of the suspected drone. k and frequency hopping sequence (period T ̂) hop ).

[0133] The jammer employs tracking jamming: during each dwell time, it transmits a signal of Gaussian white noise with the same frequency as the target signal, after bandpass filtering, with a bandwidth of 5 MHz and a power 10 dB higher than the target signal.

[0134] Synchronization method: The signal envelope separated in step S3 is used to detect the transition time, and the interference frequency is switched 0.1 ms in advance.

[0135] The formula for calculating the interference signal-to-noise ratio (JNR) is: JNR = P jam -P sig -L path .

[0136] Wherein: JNR: the ratio of the interference signal power to the target communication signal power (dB).

[0137] P jam The interference power (dBm) at the target location is calculated in step S41. P sig The original communication signal power (dBm) at the target location is obtained by back-calculating the RSSI and distance from step S3. L path Path loss difference (generally < 5 dB due to the different positions of the interfering transmitter and the target signal transmitter) requires a JNR > 10 dB to raise the bit error rate of the communication link to > 10. -3 This can lead to image transmission interruption or flight control command errors.

[0138] The purpose of step S42 is to achieve precise interference that is synchronized with the frequency hopping of the UAV.

[0139] S43 generates pseudo-satellite signals simulating the global navigation satellite system. It first simulates the real position to lock onto the target UAV, and then slowly changes the pseudorange at a speed lower than its flight control anomaly detection threshold to induce its flight trajectory to gradually deviate from the original route.

[0140] An airborne GPS simulator (an FPGA-based code baseband generator) generates C / A code pseudosatellite signals in the L1 band (1575.42 MHz) to simulate the ephemeris and pseudorange of six satellites.

[0141] "Slow drifting" strategy: The first 10 seconds: simulate a real GPS signal (consistent with the actual location of the target).

[0142] Then at 0.1 m / s 2 The false acceleration causes the calculated position to move slowly, increasing the deceptive displacement Δr(t) = [0.5t, 0, 0] every 5 seconds (horizontal direction only).

[0143] The decoy displacement speed is slower than the flight control's anomaly detection threshold (typically > 0.5 m / s). 2 (Only then will the alarm be triggered), thus going unnoticed.

[0144] The pseudorange generation formula is: ρ fake (i) (t)=∥r sat (i) -(r true (t)+Δr(t))∥+c⋅δt sv (i) +δ iono +δ trop .

[0145] Where: ρ fake (i) (t): The pseudorange (in meters) of the i-th simulated satellite.

[0146] r sat (i) : The position vector (3×1) of the i-th simulated satellite, calculated from the ephemeris parameters.

[0147] r true (t): The actual location of the drone (in meters).

[0148] Δr(t): The applied decoy displacement vector (increases linearly with t, in meters).

[0149] c = 3 × 10 8 m / s: speed of light.

[0150] δt sv (i) : Simulated satellite clock bias (s).

[0151] δ iono δ trop : Delayed model values ​​(m) for the ionosphere and troposphere.

[0152] The position calculated by the jammed drone after receiving the signal gradually deviated from the target position, and the flight controller could not distinguish between the real and fake positions, eventually leading the drone to a safe area.

[0153] The purpose of step S43 is to induce the unauthorized drone to make a forced landing or return by forging GPS signals after the communication link of the unauthorized drone is interrupted.

[0154] The purpose of step S4 is to transmit directional jamming signals or spoof GPS signals based on the location and signal characteristics of the suspicious drone determined in step S3.

[0155] Step S5: On the airborne edge computing unit, based on historical flight data and real-time obstacle avoidance and countermeasure effects, the fusion weights and path planning parameters in the adaptive weighted fusion algorithm are optimized online.

[0156] In this embodiment, step S5 may specifically include the following steps: S51 compares the obstacle prediction results with the actual perception results, calculates the short-term prediction error of each sensor, and uses the gradient descent method to update the confidence calculation parameters online, thereby reducing the fusion weight of sensors with large errors.

[0157] Define the short-term prediction error e of the i-th sensor i (k)=|z ̂ i (k)-z i (k)|, where z ̂ i (k) is the estimated value that is back-mapped to the sensor's observation space based on the EKF prediction.

[0158] Update the parameter β in step S12 using online gradient descent. i (Each sensor has its own adjustment coefficient, with an initial value of 0.7).

[0159] It updates every second, and the update step size is proportional to the error size.

[0160] The formula for updating the weight parameters is: β i (k+1)=β i (k)-η (∂e i(k) 2 ) / ∂β i ; ∂e i 2 / ∂β i ≈ 2e i (k)⋅(z ̂ i (k;β i +δβ)-z ̂ i (k;β i )) / δβ.

[0161] Where: β i (k): The adjustment parameter of the i-th sensor at the k-th update; η=0.01: Learning rate; e i (k): Prediction error (in m); δβ=0.01: Small perturbation.

[0162] If a sensor frequently malfunctions, its β value will decrease, leading to a reduction in the weight in the formula of step 1.2, thus achieving adaptive fault tolerance. For example, in heavy fog, the error of a lidar increases, and the weight automatically decreases from 0.5 to 0.2.

[0163] The purpose of step S51 is to reduce the weight of sensors with large errors when the obstacle trajectory predicted in step S2 is inconsistent with the actual perception results in step S1.

[0164] S52 uses the statistical features of the environmental point cloud and the current obstacle avoidance success rate as the state, and the weight parameters and safe distance in the path planning cost function as the actions to build a deep reinforcement learning model. The weight parameters are optimized through reward signals, enabling the UAV to adapt to different scenarios.

[0165] Using a deep Q-network (DQN), state s t This includes: point cloud average spatial frequency (voxel occupancy rate), maximum hole size, and current obstacle avoidance success rate (last 20 obstacle avoidance events).

[0166] Action a t : λ1∈{0.1,0.3,0.6}, λ2∈{0.1,0.3,0.6}, λ3=1-λ1-λ2, d safe There are 3×3×3=27 discrete actions in the range {0.8, 1.2, 1.8}m.

[0167] Reward R t Successful obstacle avoidance +10; Collision -100; Path length -0.1 for every 1 meter exceeding the straight-line distance; Average curvature -0.1 for every 0.5 rad / m exceeding the straight-line distance.

[0168] Network structure: 3 fully connected layers (128-64-27), using ReLU activation, ε-greedy exploration (ε=0.2).

[0169] The formula for updating the Q value is: Q(s t ,a t )←Q(s t ,a t )+α[R t+1 +γmax a Q(s t+1 ,a)-Q(s t ,a t )).

[0170] Where: Q(s) t ,a t ): State-action value function. α=0.2: Learning rate. γ=0.9: Discount factor. R t+1 In state s t Execute a t The immediate reward received afterward. (max) a Q(s t+1 ,a): The maximum estimated value of the next state.

[0171] After 50 flights, the system automatically selected forest mode (smaller d). safe (larger λ3) or open zone mode (larger d) safe (With a smaller λ2), the obstacle avoidance success rate increased from 90% to 98%.

[0172] The purpose of step S52 is to optimize the cost function weights λ1, λ2, λ3 and the safety distance d in step S23. safe It can adapt to different scenarios.

[0173] In the S53, multiple drones encrypt and upload the gradient parameters of the reinforcement learning model trained locally to the cloud. The cloud performs federated averaging to obtain a global model, which is then compressed into a lightweight model through knowledge distillation and distributed to each drone.

[0174] Each drone is locally trained with a DQN network. local After every 10 flights, the network parameters θ will be... local The gradient encryption is uploaded to the ground cloud server.

[0175] The server performs federated averaging (FedAvg) to generate the global network Q. global .

[0176] Using knowledge distillation: with Q global(5 layers, Teacher) output soft labels are used to train a 3-layer student network Q. student The loss function is KL divergence and cross-entropy.

[0177] The student network distributes the data to all drones after every 50 flights.

[0178] The federal average calculation formula is: .

[0179] in: n: Global model parameters in round t+1. M: Number of drones participating in training (e.g., 10). m : The number of local training samples for the m-th drone (the size of the experience pool from the most recent 10 flights). Total number of samples. : The model parameters uploaded by the m-th drone in round t.

[0180] Without revealing the flight paths of each drone, the obstacle avoidance performance is collectively improved, and the student model inference time is only 1.2ms, which is suitable for edge computing.

[0181] The purpose of step S53 is to enable multiple drones to learn collaboratively, summarize their optimized strategies, distill them into a lightweight model, and then distribute it.

[0182] The purpose of step S5 is to optimize the sensor fusion weights and planning parameters online using historical data and real-time results on the airborne edge computing unit.

[0183] Step S6: Perform multi-dimensional quantitative evaluation of the countermeasures. If the evaluation result is lower than the preset threshold, automatically adjust the countermeasures parameters and store successful or unsuccessful cases in the knowledge base for strategy migration in similar scenarios.

[0184] In this embodiment, step S6 may specifically include the following steps: S61, after a countermeasure, measures the communication interruption status of the suspected drone, the deviation angle between its actual flight trajectory and the desired trajectory, and the time from the start of the interference to the elimination of the threat, and calculates a weighted comprehensive success rate index.

[0185] The evaluation is conducted across three dimensions: Communication interruption index I comms Within 1 second after the interference, does step S3 no longer detect the frequency hopping / image transmission signal of the suspicious drone?

[0186] Trajectory deviation Δψ: Calculates the angle error between the actual trajectory and the desired trajectory of a suspicious drone by continuously tracking it with its own radar / vision.

[0187] Threat termination time T neutralize : The time (in seconds) from the start of the interference until the suspected drone descends to below 5 meters in altitude.

[0188] Overall Success Rate Index S eff The calculation formula is: S eff =w c ⋅I comms +w d ⋅min(1,|Δψ| / ψ th )+w t ⋅max(0,1-T neutralize / T max ).

[0189] Wherein: S eff : Overall success rate index, range [0,1].

[0190] I comms : Communication interruption indication, take 0 (no interruption) or 1 (interruption).

[0191] Δψ: Trajectory deviation angle error (degrees), calculated from the directional difference between the actual trajectory and the desired deceptive trajectory. th =30°: Threshold angle. T neutralize Threat termination time (s). T max =15 seconds (anything over 15 seconds is considered inefficient). w c =0.5,w d =0.3,w t =0.2: Weighting coefficient.

[0192] If S eff If the value is less than 0.6, the current countermeasure strategy is deemed invalid, and step S62 is triggered.

[0193] The purpose of step S61 is to define quantitative indicators to determine whether the interference / deception is successful.

[0194] S62 automatically increases interference transmission power or switches deception strategy when the overall success rate index is lower than the preset threshold, and adds the radio frequency fingerprint of the suspicious drone to the cloud blacklist.

[0195] If I comms If the value is 0 (communication is not interrupted), then the interference power P increases every 0.5 seconds. jam 1 dB, until the legal maximum of 30 dBm is reached.

[0196] If Δψ < 10° (deception ineffective), then switch from slow drift state to instantaneous jump and constant drift state: first apply a pseudorange offset of Δr = [30m, 0, 0] instantaneously, then at 0.2 m / s2 Continuing to drift.

[0197] If T neutralize >15s, record the radio frequency fingerprint of the suspicious drone (step S32), and upload it to the cloud blacklist for other drones to quickly identify.

[0198] The formula for power increment is: P jam (k+1)=min(P max , P jam (k)+ΔP⋅(1-S eff )).

[0199] Where: P jam (k): Interference power (dBm) at the k-th adjustment. ΔP = 0.5 dB: Step value. P max =30dBm: Legal upper limit. eff The overall success rate index calculated in step S61.

[0200] The worse the countermeasure effect (S) eff The smaller the value, the faster the power increases. When the power reaches its limit, switch to frequency hopping tracking mode (from frequency sweeping interference to matched interference).

[0201] The purpose of step S62 is to adjust the jamming power, beamwidth, or decoy speed when the evaluation finds that the countermeasure effect is poor.

[0202] S63 stores the input feature vector, the countermeasure strategy, and the overall success rate index of each countermeasure case as a record in the knowledge base. When the feature distance between a new case and an existing case is less than a preset value, the countermeasure strategy of that case is directly invoked.

[0203] Use an incremental decision tree (a variant of the C4.5 algorithm) to store cases, with the input feature vector f=[f hop ,P r [,weather,anti_jamming_flag], the output is the recommended strategy π (interference type, deception rate).

[0204] When the weighted Euclidean distance between a new case and an existing case is less than a threshold, the strategy is directly reused to avoid relearning.

[0205] The knowledge base has a capacity of 1000 entries and uses an LRU (Least Recently Used) eviction policy.

[0206] The formula for calculating feature distance is: .

[0207] Where: d: Feature distance between the new case and existing cases (dimensionless). Q: Number of features (e.g., 4). The qth feature of the new case (e.g., frequency hopping period of 0.01 s). σ is the q-th feature value of a case in the knowledge base. q : The standard deviation of the q-th feature across all cases (for normalization).

[0208] If d < 1.2, the strategy of this case will be reused. After executing countermeasures multiple times, the system can achieve an instantaneous strategy response of less than 0.2 seconds against common DJI Phantom series drones, custom frequency hopping drones, and other black-market aircraft, meeting the requirement of interference response time of less than 0.3 seconds.

[0209] The purpose of step S63 is to store historical countermeasure cases in the knowledge base so that the optimal strategy can be directly invoked when a similar scenario is encountered again.

[0210] The purpose of step S6 is to evaluate the effectiveness of the countermeasures in step S4. If the expected results are not achieved, the strategy adjustment is triggered and feedback is given to steps S3 and S4.

[0211] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0212] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0213] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0214] Example 2 Further reference Figure 2 As a response to the above Figure 1 The present invention provides an embodiment of an autonomous obstacle avoidance and countermeasure device for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, which is similar to the method shown. Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0215] like Figure 2 As shown, the UAV autonomous obstacle avoidance and countermeasure device 70 based on multi-sensor fusion described in this embodiment includes: a fusion mapping module 71, a path replanning module 72, an identity determination module 73, a cooperative countermeasure module 74, an online optimization module 75, and a self-evolution module 76. Wherein: The fusion mapping module 71 is used to synchronize the airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors in time and space, and dynamically evaluate the confidence of each sensor according to environmental changes, and use an adaptive weighted fusion algorithm to build a three-dimensional environment map in real time. The path replanning module 72 is used to identify and track dynamic obstacles based on the three-dimensional environment map, predict their short-term motion trajectory, calculate the collision probability, and replan the local obstacle avoidance path when there is a risk of collision. The identity identification module 73 is used to passively monitor the communication frequency band of drones, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of suspicious drones, and at the same time estimate their incoming wave direction and distance. The collaborative countermeasure module 74 is used to form a directional beam pointing to the suspicious drone based on the identified location and signal characteristics of the suspicious drone, transmit tracking jamming signals to cut off its communication link, and / or send deceptive navigation signals to guide it to deviate from its flight path. The online optimization module 75 is used to optimize the fusion weights and path planning parameters in the adaptive weighted fusion algorithm online on the airborne edge computing unit based on historical flight data and real-time obstacle avoidance and countermeasure effects. The self-evolution module 76 is used to perform multi-dimensional quantitative evaluation of countermeasures. If the evaluation result is lower than the preset threshold, the countermeasure parameters are automatically adjusted, and successful or unsuccessful cases are stored in the knowledge base for subsequent strategy migration in similar scenarios.

[0216] Example 3 To address the aforementioned technical problems, embodiments of the present invention also provide a computer device. Please refer to [link / reference needed]. Figure 3 , Figure 3 This is a basic structural block diagram of the computer device in this embodiment.

[0217] The aforementioned computer device 8 includes a memory 81, a processor 82, and a network interface 83 that are interconnected via a system bus. It should be noted that only the computer device 8 with components 81, 82, and 83 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0218] The aforementioned computer devices can be desktop computers, laptops, handheld computers, and cloud servers, among other computing devices. These devices can facilitate human-computer interaction with users through keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0219] The aforementioned memory 81 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the aforementioned memory 81 may be an internal storage unit of the aforementioned computer device 8, such as the hard disk or memory of the computer device 8. In other embodiments, the aforementioned memory 81 may also be an external storage device of the aforementioned computer device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 8. Of course, the aforementioned memory 81 may also include both the internal storage unit and its external storage device of the aforementioned computer device 8. In this embodiment, the aforementioned memory 81 is typically used to store the operating system and various application software installed on the aforementioned computer device 8, such as computer-readable instructions for a multi-sensor fusion-based autonomous obstacle avoidance and countermeasure method for unmanned aerial vehicles. In addition, the aforementioned memory 81 can also be used to temporarily store various types of data that have been output or will be output.

[0220] In some embodiments, the processor 82 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 82 is typically used to control the overall operation of the computer device 8. In this embodiment, the processor 82 is used to execute computer-readable instructions stored in the memory 81 or to process data, for example, to execute the computer-readable instructions of the aforementioned multi-sensor fusion-based UAV autonomous obstacle avoidance and countermeasure method.

[0221] The network interface 83 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the computer device 8 and other electronic devices.

[0222] Example 4 The present invention also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the above-described method for autonomous obstacle avoidance and countermeasure of unmanned aerial vehicles based on multi-sensor fusion.

[0223] The beneficial effects of implementing the above methods are as follows: (1) Significantly improves perception reliability and environmental adaptability: By dynamically evaluating the confidence of each sensor and adopting adaptive weighted fusion, the system can automatically reduce the weight of low-quality sensors in complex or sudden environments (such as rain, fog, strong backlight, electromagnetic interference), ensuring the continuity and accuracy of the three-dimensional environment map, and greatly enhancing the survivability of UAVs in unstructured environments.

[0224] (2) High level of intelligence and precision in obstacle avoidance and countermeasures: It can predict the short-term trajectory of dynamic obstacles and calculate the collision probability, so as to achieve proactive obstacle avoidance rather than blind abrupt turns, thereby improving flight safety and path smoothness. At the same time, through directional beam tracking jamming, it can accurately distinguish between friend and foe, and carry out surgical countermeasures against suspicious UAVs, avoiding broad-spectrum energy leakage and impact on surrounding equipment.

[0225] (3) Possesses continuous evolution and closed-loop optimization capabilities: Online optimization of historical parameters is achieved on the airborne edge computing unit, and countermeasures can be automatically adjusted based on multi-dimensional quantitative evaluation results, forming a closed loop of perception-decision-action-evaluation-optimization. After successful and failed cases are stored in the knowledge base, strategies for similar scenarios can be quickly migrated, making the UAV system more and more intelligent with use, and significantly reducing the cost of manual operation and maintenance and parameter debugging.

[0226] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0227] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles (UAVs) based on multi-sensor fusion, characterized in that, Includes the following steps: The system performs time and space synchronization of airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors, dynamically evaluates the confidence of each sensor according to environmental changes, and uses an adaptive weighted fusion algorithm to build a three-dimensional environment map in real time. Based on the three-dimensional environment map, dynamic obstacles are identified and tracked, their short-term motion trajectories are predicted, collision probability is calculated, and local obstacle avoidance paths are replanned when there is a risk of collision. Passively monitor the communication frequency band of drones, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of suspicious drones, and estimate their direction of arrival and distance. Based on the identified location and signal characteristics of the suspicious drone, a directional beam is formed pointing to the suspicious drone, a tracking jamming signal is emitted to cut off its communication link, and / or a deceptive navigation signal is sent to guide it off course. On the airborne edge computing unit, based on historical flight data and real-time obstacle avoidance and countermeasure effects, the fusion weights and path planning parameters in the adaptive weighted fusion algorithm are optimized online. The countermeasures are evaluated quantitatively from multiple dimensions. If the evaluation results are lower than the preset threshold, the countermeasure parameters are automatically adjusted, and successful or unsuccessful cases are stored in the knowledge base for strategy migration in similar scenarios in the future.

2. The method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The steps of synchronizing airborne lidar, millimeter-wave radar, binocular vision, and ultrasonic sensors in time and space, dynamically evaluating the confidence level of each sensor according to environmental changes, and constructing a 3D environment map in real time using an adaptive weighted fusion algorithm specifically include: The data acquisition by each sensor is triggered by the timing signal of the Global Navigation Satellite System, and cubic spline interpolation is performed on the measurement values ​​of the low sampling rate sensors to achieve coarse time synchronization; spatial synchronization is achieved by projecting all sensor data onto a unified world coordinate system through a pre-calibrated rotation matrix and translation vector. The historical error variance and current signal-to-noise ratio of each sensor are calculated in real time. The adaptive fusion weight of each sensor is calculated based on the exponential weighting method. The measured values ​​of each sensor are weighted and averaged according to the weight to obtain the fused environmental perception data. The fused point cloud data is mapped to an octree structure, and the occupancy probability of each voxel is updated using a Bayesian filter in the form of log probability ratio, forming a three-dimensional occupancy probability map.

3. The method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The steps of identifying and tracking dynamic obstacles based on the three-dimensional environment map, predicting their short-term motion trajectory, calculating the collision probability, and replanning local obstacle avoidance paths when there is a risk of collision specifically include: The drone's own motion is removed by point cloud registration, and the remaining motion vectors are clustered to obtain dynamic obstacles. Then, the position, velocity and acceleration of each dynamic obstacle are estimated and predicted by extended Kalman filtering. Calculate the relative position vector and relative velocity vector between the drone and the dynamic obstacle, determine whether there is a collision risk based on the collision cone condition, and calculate the collision time. When the collision time is lower than the safety threshold, mark the obstacle as a dangerous obstacle. The local path is represented by a high-order Bézier curve. A cost function is constructed using the path length, curvature, and safe distance from obstacles. A new path that satisfies the UAV dynamics constraints and avoids the dangerous obstacles is obtained through optimization.

4. The method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The steps of passively monitoring the communication frequency band of drones, separating suspicious signals from the electromagnetic background, extracting radio frequency fingerprints and comparing them with a preset whitelist to identify the identity of the suspicious drone, and simultaneously estimating its direction of arrival and distance specifically include: The power spectrum is obtained by performing a fast Fourier transform on the received broadband signal. Suspicious frequency points are extracted by constant false alarm rate detection. Independent component analysis is performed on the baseband signal at the suspicious frequency points to separate the signal components from different UAV sources from the mixed signal. The frequency hopping period, modulation method, and demodulated data frame address are extracted from the separated signal components and compared with the built-in drone radio frequency fingerprint database and local whitelist. If they cannot match and the flight behavior is abnormal, the drone is identified as suspicious. The signal is received using an airborne array antenna. The direction of arrival of the suspicious UAV is estimated by a subspace decomposition algorithm. The distance is estimated based on the received signal strength and the free space propagation model.

5. The method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The steps of forming a directional beam pointing at the suspicious drone based on the identified location and signal characteristics, transmitting a tracking jamming signal to cut off its communication link, and / or sending a deceptive navigation signal to guide it off course specifically include: Based on the estimated direction of arrival, the phase offset of each transmission channel of the phased array antenna is calculated so that the main lobe of the synthesized beam is pointed at the suspected UAV, and the transmission power is adjusted according to the estimated distance so that the strength of the interference signal received at the target reaches a preset threshold. Real-time tracking of the frequency hopping or fixed operating frequency of suspicious drones; during each dwell time, transmitting a narrowband jamming signal with the same frequency and superior power as the target signal, causing the bit error rate of its communication link to rise above the interruption threshold. The system generates pseudo-satellite signals that simulate the global navigation satellite system. It first simulates the real position to lock onto the target UAV, and then slowly changes the pseudorange at a speed lower than the detection threshold of its flight control anomaly, inducing its flight trajectory to gradually deviate from the original route.

6. The method for autonomous obstacle avoidance and countermeasures by unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The step of optimizing the fusion weights and path planning parameters in the adaptive weighted fusion algorithm online on the airborne edge computing unit based on historical flight data and real-time obstacle avoidance and countermeasure effects specifically includes: The obstacle prediction results are compared with the actual perception results, the short-term prediction error of each sensor is calculated, and the confidence calculation parameters are updated online using the gradient descent method to reduce the fusion weight of sensors with large errors. Using the statistical features of environmental point clouds and the current obstacle avoidance success rate as the state, and the weight parameters and safe distance in the path planning cost function as the actions, a deep reinforcement learning model is constructed. The weight parameters are optimized through reward signals to enable the UAV to adapt to different scenarios. Multiple drones encrypted the gradient parameters of the reinforcement learning model trained locally and uploaded them to the cloud. The cloud then performed federated averaging to obtain a global model, which was then compressed into a lightweight model through knowledge distillation and distributed to each drone.

7. The method for autonomous obstacle avoidance and countermeasure of unmanned aerial vehicles based on multi-sensor fusion according to any one of claims 1 to 6, characterized in that, The steps of conducting multi-dimensional quantitative evaluation of countermeasures, automatically adjusting countermeasure parameters if the evaluation result is lower than a preset threshold, and storing successful or unsuccessful cases in a knowledge base for subsequent strategy migration in similar scenarios specifically include: After the countermeasures, the communication interruption status of the suspected drone, the deviation angle between the actual flight trajectory and the desired trajectory, and the time from the start of the interference to the elimination of the threat are measured, and a weighted comprehensive success rate index is calculated. When the overall success rate index is lower than a preset threshold, the interference transmission power is automatically increased or the deception strategy is switched, and the radio frequency fingerprint of the suspicious drone is added to the cloud blacklist. The input feature vector, the countermeasure strategy, and the comprehensive success rate index of each countermeasure case are stored as a record in the knowledge base. When the feature distance between a new case and an existing case is less than a preset value, the countermeasure strategy of that case is directly invoked.

8. A drone autonomous obstacle avoidance and countermeasure device based on multi-sensor fusion, characterized in that, include: The fusion mapping module is used to synchronize airborne lidar, millimeter-wave radar, binocular vision and ultrasonic sensors in time and space, and dynamically evaluate the confidence of each sensor according to environmental changes. It uses an adaptive weighted fusion algorithm to build a three-dimensional environment map in real time. The path replanning module is used to identify and track dynamic obstacles based on the three-dimensional environment map, predict their short-term motion trajectory, calculate the collision probability, and replan the local obstacle avoidance path when there is a risk of collision. The identity identification module is used to passively monitor the communication frequency band of drones, separate suspicious signals from the electromagnetic background, extract radio frequency fingerprints and compare them with a preset whitelist to identify the identity of suspicious drones, and at the same time estimate their incoming wave direction and distance. The collaborative countermeasure module is used to form a directional beam pointing at the suspicious drone based on the identified location and signal characteristics of the suspicious drone, transmit tracking jamming signals to cut off its communication link, and / or send deceptive navigation signals to guide it off course. The online optimization module is used to optimize the fusion weights and path planning parameters in the adaptive weighted fusion algorithm online on the airborne edge computing unit based on historical flight data and real-time obstacle avoidance and countermeasure effects. The self-evolution module is used to perform multi-dimensional quantitative evaluation of countermeasures. If the evaluation result is lower than the preset threshold, the countermeasure parameters will be automatically adjusted, and successful or unsuccessful cases will be stored in the knowledge base for strategy migration in similar scenarios in the future.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the autonomous obstacle avoidance and countermeasure method for unmanned aerial vehicles based on multi-sensor fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the UAV autonomous obstacle avoidance and countermeasure method based on multi-sensor fusion as described in any one of claims 1 to 7.

Images