A low-altitude unmanned aerial vehicle management and control method and device based on multi-modal data fusion

Abstract

This application relates to a method and apparatus for controlling low-altitude unmanned aerial vehicles (UAVs) based on multimodal data fusion, belonging to the field of computer technology. The method includes: responding to a warning event indicating the detection of a target UAV within a target airspace, acquiring multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace; fusing and analyzing the multi-source detection feature data to obtain dynamic state parameters and communication link status; constructing a crash dispersion spatial model of the target UAV based on the dynamic state parameters and wind field data from the environmental geographic feature data; performing collision detection processing on the environmental geographic feature data based on the crash dispersion spatial model to obtain a collateral damage index; and generating graded countermeasure commands for the target UAV based on the communication link status and the collateral damage index. This application has the effect of accurately predicting the target crash area and quantitatively adjusting the countermeasure intensity, thereby reducing collateral damage during the control process.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method and apparatus for the management and control of low-altitude unmanned aerial vehicles based on multimodal data fusion. Background Technology

[0002] Low-altitude drone management is a security system used to monitor, identify, and counter drones that illegally intrude or fly in violation of regulations.

[0003] In related technologies, in the safety management and control scenario of low-altitude target airspace, the direct blocking method is usually used to counter drones: after the detection system (e.g., radar or photoelectric detector) detects the intrusion warning of the target drone, the control system directly activates the countermeasures equipment, uses unified electromagnetic suppression means (e.g., maximum power) to cut off the drone's telemetry and control link or navigation signal, or directly adopts physical net capture means to force the target drone to land on the spot or crash.

[0004] Regarding the aforementioned technologies, the control equipment, when implementing countermeasures such as signal jamming, does not fully consider the fall evolution process of the target drone after it loses control and the complexity of the ground environment. For example, at the moment a drone fails due to signal jamming, its initial dynamic state (e.g., flight speed, altitude, attitude) and meteorological environmental factors within the target airspace (e.g., wind speed, wind direction) will disturb the drone's fall trajectory. Because the fall process is random and cannot be effectively predicted, using countermeasures of constant intensity, lacking dynamic adjustment of the countermeasure strategy, may cause the out-of-control drone to veer off course and fall into sensitive ground areas such as densely populated areas, transportation hubs, or hazardous chemical storage areas. This not only fails to guarantee the original intention of low-altitude security but may also cause mechanical and physical collisions, leading to uncontrollable collateral damage, indicating room for improvement. Summary of the Invention

[0005] To overcome the ground safety risks caused by the unpredictable trajectory of unmanned aerial vehicles (UAVs) that can not be controlled, this application provides a low-altitude UAV management and control method based on multimodal data fusion.

[0006] Firstly, this application provides a low-altitude unmanned aerial vehicle (UAV) control method based on multimodal data fusion, employing the following technical solution: In response to a warning event that a target UAV is detected in the target airspace, multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace are acquired. The multi-source detection feature data is fused and analyzed to obtain the dynamic state parameters and communication link status of the target UAV. Based on the dynamic state parameters and wind field data in the environmental geographic feature data, a crash dispersion spatial model of the target UAV under signal interference conditions is constructed. Based on the crash dispersion spatial model, collision detection processing is performed on the sensitive area data in the environmental geographic feature data to obtain a collateral damage index. Based on the communication link status and the collateral damage index, a graded countermeasure command is generated for the target UAV.

[0007] Secondly, this application provides a low-altitude unmanned aerial vehicle (UAV) control device based on multimodal data fusion, employing the following technical solution: The acquisition module is used to acquire multi-source detection feature data and environmental geographic feature data; the memory is used to store the program of the above-mentioned low-altitude UAV management method based on multimodal data fusion; the processor is used to load and execute the program in the memory and implement the above-mentioned low-altitude UAV management method based on multimodal data fusion.

[0008] The various embodiments disclosed above have the following beneficial effects: The low-altitude UAV control method based on multimodal data fusion, as described in some embodiments of this disclosure, can achieve accurate probability prediction of the crash trajectory of out-of-control UAVs and quantitative assessment of the risk of secondary ground damage. This solves the ground safety hazards caused by the uncertainty of the crash during countermeasures, achieving a balance between control effectiveness and ground safety. Specifically, the high risk of secondary disasters related to low-altitude control and the blind nature of countermeasures are due to the fact that existing technologies often simplify UAVs to ideal linear crash points when implementing countermeasures, without considering the nonlinear dynamic failure evolution process of the target under the influence of complex meteorological wind fields. They cannot distinguish the attribute differences between densely populated areas (e.g., schools, hospitals) and low-risk areas (e.g., green spaces, water bodies), leading to blind countermeasure decisions. Based on this, the low-altitude UAV control method based on multimodal data fusion of some embodiments of this disclosure first, in response to a warning event detecting a target UAV in the target airspace, acquires multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace. This ensures that the system can grasp the electromagnetic and physical characteristics of the UAV before countermeasures, as well as identify the geographical constraints of the UAV's location (e.g., schools, hospitals), providing detailed data input for subsequent simulations. Secondly, the multi-source detection feature data is fused and analyzed to obtain the dynamic state parameters and communication link status of the target UAV. This enables deep perception of the target's real-time flight inertia and controllability. In real-world scenarios, it can distinguish whether the target is in a stable hovering or high-speed dive state, and assess the anti-interference strength of the remote control link, providing a physical basis for determining the countermeasure entry point. Then, based on the dynamic state parameters and wind field data from the environmental geographic feature data, a spatial model of the target UAV's fall dispersion under signal interference conditions is constructed. This transforms traditional single-landing-point prediction into a three-dimensional spatial envelope with probability distribution characteristics. In actual urban low-altitude scenarios, considering wind fields between high-rise buildings (e.g., through drafts) and the UAV's roll after loss of control, the dynamic drift path of the UAV after interference can be reconstructed, reducing prediction bias and improving the robustness of prediction results in complex weather conditions. Furthermore, based on the aforementioned crash dispersion spatial model, collision detection processing is performed on the sensitive area data in the aforementioned environmental geographic feature data to obtain a collateral damage index. Therefore, by spatially overlaying the predicted crash probability distribution with sensitive ground targets (e.g., schools, hospitals, and oil depots), the potential social impact and material (e.g., building materials) loss induced by countermeasures can be assessed in advance. Finally, based on the aforementioned communication link status and the aforementioned collateral damage index, tiered countermeasure commands are generated for the aforementioned target UAVs. This enables on-demand output of control commands. In actual countermeasures, if the predicted impact area is determined to have an extremely high risk of damage, the system can adopt flexible measures such as reducing interference power and guiding relocation.If the risk is within a controllable range, a forced landing is executed. This avoids administrative collateral damage caused by interference from a single execution method. Furthermore, this embodiment, through deep decoupling and reconstruction of multimodal data, digitally monitors the out-of-control evolution process, improving the scientific nature of decision-making and the safety of execution in handling sudden drone intrusion incidents by the low-altitude security system, and providing reliable technical support for the routine management and control of drones in complex urban environments. Attached Figure Description

[0009] Figure 1 This is a flowchart of some embodiments of the low-altitude unmanned aerial vehicle (UAV) management method based on multimodal data fusion according to the present disclosure. Detailed Implementation

[0010] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0011] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0012] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0013] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0014] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0015] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] refer to Figure 1 The diagram illustrates a flowchart 100 of some embodiments of a low-altitude unmanned aerial vehicle (UAV) control method based on multimodal data fusion according to the present disclosure. This low-altitude UAV control method based on multimodal data fusion includes the following steps: Step 101: In response to an early warning event that a target UAV is detected in the target airspace, acquire multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace.

[0017] In some embodiments, the executing entity (e.g., an electronic device) of the above-described low-altitude UAV management method based on multimodal data fusion can be hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed in the hardware devices listed above. It can be implemented as multiple software programs or software modules to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0018] In some embodiments, the aforementioned implementing entity may, in response to an early warning event detecting a target UAV within the target airspace, acquire multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace. The target airspace may be a pre-defined three-dimensional geographic spatial area requiring low-altitude safety control. The target UAV may be a small aircraft (e.g., a rotorcraft, fixed-wing aircraft, or flapping-wing aircraft) that has entered the target airspace without authorization. The early warning event may be a trigger signal generated by the detection equipment indicating the intrusion of an intrusive flying object. The multi-source detection feature data may be multi-dimensional digital information describing the physical, electromagnetic, and behavioral attributes of the target UAV. The environmental geographic feature data may be comprehensive spatial information describing the natural topography, man-made structures, and meteorological conditions within the target airspace.

[0019] In some optional implementations of certain embodiments, the execution entity may, in response to a warning event that a target UAV is detected in the target airspace, acquire the multi-source detection feature data of the target UAV and the environmental geographic feature data of the target airspace, which may include the following steps: The first step involves using a heterogeneous array of sensors deployed around the target airspace to collect electromagnetic scattering characteristics, radio frequency communication characteristics, and photoelectric imaging characteristics of the target UAV in parallel, in order to obtain multi-source detection characteristic data.

[0020] The aforementioned heterogeneous detection sensor array can be a collection of sensors with various operating principles (e.g., radar, radio monitoring equipment, photoelectric cameras). The aforementioned electromagnetic scattering characteristics can be radar cross-section data formed after the target UAV reflects radar waves. The aforementioned radio frequency communication characteristics can be the radio signal spectrum characteristics of the interaction between the target UAV and its remote controller or navigation satellite. The aforementioned photoelectric imaging characteristics can be pixel array data describing the shape, color, and texture of the target UAV.

[0021] In practice, the radar antenna rotation, radio antenna scanning, and photoelectric turntable pointing can be driven in parallel through the control interface. The captured analog physical signals are converted into a computer-processable digital sequence via an analog-to-digital converter, resulting in multi-source detection feature data. For example, Doppler radar can be used to obtain the speed characteristics of a target UAV, a radio spectrum analyzer can be used to obtain the center frequency of the image transmission signal, and a high-definition camera can be used to obtain the aircraft's outline image.

[0022] The second step involves determining, based on a pre-defined Geographic Information System (GIS), a digital elevation model (DEM), three-dimensional ground feature vector data, and a pre-defined sensitive target distribution map that match the spatial coordinates of the aforementioned multi-source detection feature data. The GIS can be a computer database system used for collecting, storing, managing, and analyzing geospatial data. The spatial coordinates can be the longitude, latitude, and altitude of the target UAV in three-dimensional space. The DEM can be regular grid data describing the undulating terrain. The three-dimensional ground feature vector data can be a set of vector coordinate points representing the geometric contours of ground obstacles (e.g., buildings, vegetation, utility poles). The pre-defined sensitive target distribution map can be a thematic map marking the boundaries of high-risk prevention areas (e.g., schools, hospitals, oil depots).

[0023] In practice, spatial coordinates can be used as index keys to perform correlated queries in the spatial database of a geographic information system, extracting terrain and feature information within a preset radius of those coordinates. For example, one can query whether there are tall buildings or substations currently below the target drone.

[0024] The third step involves collecting wind speed, wind direction, and air density vector information at the corresponding altitude levels within the target airspace, based on meteorological monitoring nodes. These meteorological monitoring nodes can be sensing units such as ultrasonic anemometers, barometers, etc., deployed within or around the target airspace. The altitude levels refer to the current vertical altitude range of the target UAV. The air density vector is a physical vector describing the mass of air per unit volume and its influence from air pressure and temperature.

[0025] In practice, the measurement values ​​from meteorological monitoring nodes can be read in real time via wired or wireless communication, and the aforementioned wind speed information can be corrected in real time based on the altitude of the target UAV. For example, when the altitude level is 100 meters, the instantaneous wind speed and wind direction at that altitude can be collected.

[0026] The fourth step involves spatially overlaying the aforementioned digital elevation model, the aforementioned three-dimensional ground feature vector data, the aforementioned preset sensitive target distribution map, and the aforementioned wind speed information, wind direction information, and air density vector to obtain environmental geographic feature data. This spatial overlay can be achieved using computer graphics technology to unify geographic and meteorological data from different sources into a single three-dimensional geographic coordinate system for fusion processing.

[0027] In practice, firstly, the target spatial domain can be discretized in three dimensions using the octree algorithm to generate a spatial voxel set. Next, using a digital elevation model as the spatial reference, three-dimensional land feature vector data is linked through spatial indexing, and each voxel is labeled with physical collision attributes and sensitivity levels. Then, a bilinear interpolation algorithm is used to inject discrete wind field data into the attribute fields of the voxels. Finally, each voxel encapsulates a multi-dimensional attribute operator containing terrain, obstacles, and dynamic wind force vectors, forming a dynamic holographic map with environmental disturbance attributes, thus obtaining environmental geographic feature data.

[0028] Step 102: Perform fusion and analysis processing on the multi-source detection feature data to obtain the dynamic state parameters and communication link status of the target UAV.

[0029] In some embodiments, the aforementioned execution entity can perform fusion analysis on the aforementioned multi-source detection feature data to obtain the dynamic state parameters and communication link status of the target UAV. The fusion analysis can be a process that utilizes computing power to perform spatial alignment, temporal synchronization, and physical quantity extraction on detection information from multiple heterogeneous sensors. The aforementioned dynamic state parameters can be a set of physical indicators describing the macroscopic motion properties of the target UAV in three-dimensional space, and may include, but are not limited to, real-time position coordinates, velocity vectors, and flight attitude. The aforementioned real-time position coordinates can be numerical values ​​of the target in latitude, longitude, and altitude dimensions. The aforementioned velocity vector can be the magnitude and direction of the target's displacement per unit time, and can be a combination of horizontal and vertical velocities. The aforementioned flight attitude can be a rotation angle describing the target UAV's body relative to a reference coordinate system, and may include, but is not limited to, pitch angle, roll angle, and yaw angle. The aforementioned communication link status can be a set of features characterizing the quality and protocol type of the radio signal connection between the target UAV and its control terminal (e.g., a remote controller), and may include, but is not limited to, signal-to-noise ratio, received signal strength indication, and the communication protocol. The aforementioned communication protocol can be a data transmission and control logic specification agreed upon between the UAV and the controller.

[0030] In practice, firstly, by loading a preset coordinate transformation matrix, the radar polar coordinates and photoelectric pixel coordinates in the multi-source detection feature data can be mapped to a unified geodetic coordinate system. Then, a linear interpolation algorithm is used to align data from different sampling frequencies to the same millisecond-level time point on the time axis. Next, the spatiotemporally aligned data is input as observation variables into a preset extended Kalman filter model. Through recursive calculations using the state transition equation, random noise from the sensors is filtered out, yielding the dynamic state parameters of the target UAV. Then, a fast Fourier transform is performed on the radio frequency signals in the multi-source detection feature data to extract the signal's center frequency and bandwidth characteristics. Based on a pre-stored protocol feature fingerprint database, the current communication link status of the target UAV is determined. The aforementioned geodetic coordinate system is a three-dimensional rectangular coordinate system established with the Earth's center of mass as the origin. The aforementioned coordinate transformation matrix can be a numerical array representing the translation vector and rotation parameters (e.g., Euler angles) of the photoelectric sensor relative to the geodetic coordinate system, as well as internal parameters such as the camera focal length, and can be used to achieve projection mapping from pixel two-dimensional space to real three-dimensional space. The aforementioned linear interpolation algorithm can be a computational method for estimating the value of unknown points by constructing a linear function based on known point data. The aforementioned extended Kalman filter model can be a mathematical model for optimal state estimation of nonlinear systems. The aforementioned fast Fourier transform can be a computational algorithm for converting time-domain signals into frequency-domain components. The aforementioned protocol feature fingerprint database can be a pre-collected and stored database of spectral feature mappings for various UAV communication protocols.

[0031] For example, after acquiring the range and azimuth features provided by the radar (sampled at time T1) and the angle features provided by the electro-optical camera (sampled at time T2), a linear interpolation algorithm is used to calculate the simulated angle of the electro-optical camera at time T1, thus matching the two in the spatiotemporal dimension. Next, using an extended Kalman filter model combined with the UAV's inertial motion equations, the current speed of the target UAV is calculated to be 15 m / s, its altitude to be 120 m, and its yaw angle to be 30 degrees, obtaining the dynamic state parameters. Finally, the radio frequency signal is analyzed, revealing a 2.4 GHz frequency hopping characteristic and a bandwidth of 20 MHz. This matches the protocol to a mainstream civilian UAV protocol, and its signal-to-noise ratio is identified as 18 dB, thus determining that the communication link status can be "good signal and successful link protocol identification."

[0032] Step 103: Based on the dynamic state parameters and wind field data in the environmental geographic feature data, construct a spatial model of the target UAV's crash dispersion under signal interference conditions.

[0033] In some embodiments, the aforementioned executing entity can construct a crash dispersion spatial model of the target UAV under signal interference conditions based on the aforementioned dynamic state parameters and wind field data in the aforementioned environmental geographic feature data. The crash dispersion spatial model can be a three-dimensional geometric probability envelope characterizing the possible crash trajectory and impact range of the target UAV under the coupling effects of gravity, aerodynamics, and random wind fields after losing effective control. The aforementioned signal interference conditions can be a preset operating state in which the target UAV is subjected to countermeasures such as electromagnetic suppression, protocol spoofing, or navigation hijacking, resulting in the complete or partial failure of its telemetry and control link or navigation link.

[0034] In practice, dynamic state parameters reflecting the physical state of the target UAV at the moment of loss of control can be extracted and wind field data can be imported. Then, numerical simulation can be used to predict the spatial displacement evolution of the target UAV under gravitational acceleration and wind disturbance, thereby determining the risk diffusion boundary.

[0035] For example, in a scenario where the target drone is flying at high speed and encountering strong crosswinds, a funnel-shaped three-dimensional region covering all possible crash sites can be constructed by calculating the inertial glide distance and wind deflection after loss of control, thus obtaining a crash dispersion spatial model.

[0036] In some optional implementations of certain embodiments, the execution entity may construct a spatial model of the target UAV's crash dispersion under signal interference conditions based on the aforementioned dynamic state parameters and wind field data from the aforementioned environmental geographic feature data. This may include the following steps: The first step involves performing parameter analysis on the aforementioned dynamic state parameters to obtain the initial spatial coordinates, initial motion vector, and flight attitude information of the target UAV. The initial spatial coordinates can be the latitude, longitude, and altitude values ​​of the target UAV during countermeasures. The initial motion vector can be the three-dimensional velocity vector of the target UAV when it is out of control, and can include the magnitude and direction of the velocity. The flight attitude information can be the rotation angles of the target UAV relative to the horizontal plane and geographic north, and can include pitch, roll, and yaw angles.

[0037] In practice, dynamic state parameter data packets can be read from memory and decomposed according to a preset communication protocol field format, converting the binary data stream into vector data. For example, the initial spatial coordinates are (116.39, 39.91, 100m), the initial motion vector is 10m / s eastward, and the pitch angle in the flight attitude information is 5 degrees.

[0038] The second step involves performing failure dynamics analysis on the initial spatial coordinates, initial motion vectors, and flight attitude information, based on a preset signal interference failure mode, to obtain the baseline fall trajectory equation. The signal interference failure mode can be a pre-defined physical scenario describing the motor speed response and aerodynamic layout changes of the target UAV after signal interruption. Examples include "full motor shutdown forced landing" or "constant power drift." The failure dynamics analysis can be a calculation process using Newton's laws of motion and fluid dynamics equations to simulate the trajectory of an object under the influence of only gravity and basic air resistance. The baseline fall trajectory equation can be a spatial position function describing the change of the target UAV's center of gravity over time under ideal conditions without external random wind interference.

[0039] In practice, a six-degree-of-freedom (6-DOF) system of motion differential equations can be constructed based on the selected signal interference failure mode. The initial spatial coordinates, initial motion vectors, and flight attitude information are then used as initial values ​​for numerical integration to solve the differential equations. For example, in the "motor stop" mode, the air drag coefficient and fuselage mass are substituted into the projectile motion equations to obtain a smooth parabolic curve, i.e., the baseline fall trajectory equation. In the "constant power drift" mode, the residual motor speed is converted into a thrust vector, which, along with gravity and air drag, is input as an external force term into the six-DOF system of motion differential equations for solution.

[0040] The third step is to determine the wind field data within the aforementioned environmental geographic feature data. This wind field data can be a set of vector data describing the airflow patterns within the target airspace.

[0041] Fourth, based on the aforementioned wind field data, a multi-step Monte Carlo perturbation simulation is performed on the aforementioned baseline fall trajectory equation to obtain a probabilistic fall trajectory set. This multi-step Monte Carlo perturbation simulation is a numerical calculation method that uses a large number of random samples to calculate the output distribution of an uncertain system. The probabilistic fall trajectory set can be a collection of simulated fall paths with multiple random perturbation characteristics generated in multiple simulations.

[0042] In practice, wind speed and direction from wind field data can be used as the mean, with pre-set Gaussian random noise added as a perturbation term, to conduct several (thousands to tens of thousands) independent simulation experiments. In each experiment, the wind speed vector with random perturbation is vector-synthesized with the current velocity of the target UAV to calculate the relative airflow velocity, thereby deriving the corresponding aerodynamic drag and torque (i.e., wind force operator), which is then injected into the step size calculation of the baseline fall trajectory equation, and each path is recorded. For example, performing 10,000 simulations, with wind speeds randomly fluctuating between 4 m / s and 6 m / s in each simulation, yields 10,000 flight paths of varying shapes, collectively forming a probabilistic fall trajectory set.

[0043] The fifth step involves performing a three-dimensional spatial envelope fitting process on each probability fall trajectory in the aforementioned probability fall trajectory set to obtain a fall scattering spatial model. This three-dimensional spatial envelope fitting process can be achieved by using computational geometry algorithms to extract the outermost boundary or probability density distribution limit of a set of discrete spatial paths.

[0044] In practice, the outer shell of all discrete coordinate points in the probabilistic crash trajectory set can be extracted, and a three-dimensional closed entity can be generated using the convex hull algorithm (or concave hull algorithm). For example, by voxelizing and aggregating 10,000 paths, the path coverage at a 95% confidence level can be calculated, generating a geometry resembling an inclined frustum of a cone. This geometry can serve as a crash scattering spatial model, characterizing the various spatial ranges where a drone might crash.

[0045] Step 104: Based on the fall scattering spatial model, collision detection processing is performed on the sensitive area data in the environmental geographic feature data to obtain the incidental damage index.

[0046] In some embodiments, the aforementioned executing entity can perform collision detection processing on sensitive area data in the aforementioned environmental geographic feature data based on the aforementioned fall dispersion spatial model to obtain a collateral damage index. The aforementioned sensitive area data can be a set of geographic entity attributes (e.g., gas stations, schools, hospitals) pre-labeled in the environmental geographic feature data that have extremely low tolerance for mechanical impact or secondary disasters. The aforementioned collision detection processing can be a calculation process that uses computer graphics algorithms to determine whether two or more three-dimensional geometric models geometrically intersect or overlap in virtual space. The aforementioned collateral damage index can be a numerical rating standard used to quantitatively assess the degree of damage caused to ground personnel, buildings, or social facilities after the target drone crashes.

[0047] As an example, collision detection can be used to determine that the crash scattering spatial model overlaps with a school (i.e., sensitive area data). By combining the drone's mass and speed, it can be calculated that if the drone crashes there, it will cause extremely high casualties, thus outputting a collateral damage index of "95" (out of 100).

[0048] In some optional implementations of certain embodiments, the execution entity may perform collision detection processing on sensitive area data in the aforementioned environmental geographic feature data based on the aforementioned fall dispersion spatial model to obtain the incidental damage index, which may include the following steps: The first step involves vectorizing the sensitive area data within the aforementioned environmental geographic feature data to extract its spatial contours. This vectorized boundary extraction process can utilize edge detection or contour tracking algorithms to transform raster images or point cloud data into a vector data structure composed of vertex coordinates and polygon edges. The resulting spatial contours of the sensitive areas can be closed polygons or polyhedra describing the geometric shell of a specific sensitive target in a three-dimensional Earth coordinate system.

[0049] In practice, environmental geographic feature data can be parsed. If sensitive area data exists in voxel form (each voxel carries a sensitive attribute label), a connected component labeling algorithm (such as 26-neighbor seed filling) is used to merge adjacent sensitive voxels into independent target objects. For each target object, the projection point set of all its voxels on the horizontal plane (XY plane) is extracted, and the convex hull or concave hull of this point set is calculated (using the Alpha Shapes algorithm, setting the radius parameter). The coordinates of the vertices of the base polygon are obtained by measuring the height in meters. The building height is taken as the difference between the maximum and minimum Z coordinates of the voxels in the target object. Finally, the base polygon is stretched upwards along the Z-axis by a 3D stretching algorithm to generate a closed geometric bounding box (a cuboid for a rectangular base and a prism for an arbitrary polygon base), which is the spatial outline of the sensitive area. For example, when the geographical information of a school is read, the latitude and longitude of its four corners (north, south, east, and west) are extracted and combined with the building height of 15 meters to generate a cuboid spatial outline of the sensitive area.

[0050] The second step is to perform a three-dimensional spatial intersection calculation on the above-mentioned fall dispersion spatial model and the above-mentioned sensitive area spatial contour to obtain the overlap range of the target UAV's fall area.

[0051] The aforementioned three-dimensional spatial intersection operation can be a mathematical geometric calculation method that uses Boolean operations to solve for the common part of two three-dimensional geometric shapes. The aforementioned overlapping range of the falling area can be the intersection geometry jointly occupied by the falling scattering spatial model and the sensitive area spatial contour in three-dimensional space.

[0052] In practice, a 3D computational geometry engine (e.g., the Boolean operation module in CGAL or Libigl) can be invoked to calculate the intersection lines between the faces of the fall distribution spatial model and the faces of the sensitive area spatial contour, thereby extracting the common volume and obtaining the fall area overlap range. For example, when the lower half of the funnel-shaped fall distribution spatial model crosses the sensitive area spatial contour of the school, the computer solves for an irregular polyhedron at the intersection of the two, meaning that this irregular polyhedron can be the fall area overlap range.

[0053] The third step involves performing kinetic energy conversion processing on the aforementioned dynamic state parameters within the overlapping area of ​​the impact zone to obtain the mechanical impact kinetic energy at the expected moment of collision. This kinetic energy conversion processing can be based on classical mechanics formulas (…). This refers to the process of converting the mass and instantaneous velocity of an object into joules of energy. The aforementioned mechanical impact kinetic energy can be considered as the total physical energy that the target UAV can do when it contacts the top surface of the overlapping area of ​​the impact zone.

[0054] In practice, the static mass data of the target UAV can be extracted from its factory specifications. Combined with the combined terminal velocity of the vertical and horizontal velocities at the altitude corresponding to the overlapping area of ​​the impact zone, from the dynamic state parameters, this data can be substituted into the kinetic energy formula to obtain the mechanical impact kinetic energy. For example, for a UAV with a mass of 10 kg, the combined terminal velocity when falling to the boundary of the overlapping impact zone is 30 m / s, resulting in a mechanical impact kinetic energy of 4500 joules.

[0055] The fourth step involves performing a level mapping process on the geographic element attributes within the overlapping areas to obtain sensitivity level weights. These geographic element attributes can be textual information describing ground social attributes (e.g., target use, population density, and hazardous chemical reserves). The level mapping process can be performed by querying a pre-defined mapping dictionary to convert the textual descriptions into quantitative numerical coefficients. These sensitivity level weights can characterize the severity of secondary disasters caused when different ground targets suffer the same level of impact. These sensitivity level weights support dynamic adjustment based on real-time perceived time or population heatmaps. For example, when a school is detected to be in session, its corresponding risk baseline bias can be dynamically increased.

[0056] In practice, the land cover classification labels corresponding to the overlapping areas can be read, and the floating-point values ​​corresponding to the labels can be retrieved in a pre-set weighted knowledge base (e.g., "school" = 1.0, "hospital" = 1.5, "chemical plant" = 2.0, "abandoned warehouse" = 0.2) to obtain the sensitivity level weight.

[0057] For example, if the geographic feature attribute of this area is "outpatient building of a tertiary hospital", the sensitivity level weight corresponding to this attribute can be found to be "1.5" through the mapping table. If it is "abandoned warehouse", the corresponding sensitivity level weight is "0.2".

[0058] The fifth step involves quantifying the damage based on the overlapping areas of the impact zones, considering the mechanical impact kinetic energy and sensitivity level weights, to obtain the incidental damage index. This damage quantification process can be a workflow that integrates physical impact energy, ground vulnerability, and spatial distribution, using mathematical modeling to derive risk assessment indicators.

[0059] In practice, the overlapping area of ​​the impact zone can be subdivided, and risk assessment can be performed on each subdivided unit using probabilistic and energetic models. Social attribute characteristics can be introduced for weighted calculation to obtain the collateral damage index. For example, spatial calculus can be used to calculate the collateral damage index of 88 (which belongs to extremely high risk) by combining the mechanical impact kinetic energy of 4500 joules, the sensitivity level weight of 1.5 for hospitals, and the probability of the overlapping area of ​​the impact zone.

[0060] In some optional implementations of certain embodiments, the execution entity may perform damage quantification processing on the mechanical impact kinetic energy and the sensitivity level weight based on the overlapping range of the impact area to obtain the incidental damage index, which may include the following steps: The first step is to spatially discretize the overlapping area of ​​the aforementioned landing region to obtain a three-dimensional spatial voxel set. This spatial discretization process can be a procedure of dividing a continuous three-dimensional geometric space into several uniform, non-overlapping cubic micro-elements (voxels) according to a preset resolution. The aforementioned three-dimensional spatial voxel set can be the collection of all the tiny voxels constituting the overlapping area of ​​the landing region.

[0061] In practice, a spatial resolution step of 1 meter × 1 meter × 1 meter can be used to divide the irregular overlapping area into regular cubic units using the Octree partitioning algorithm.

[0062] For example, for an overlapping region with a volume of 500 cubic meters, it is cut and encoded to generate a standard voxel set with 500 sides of 1 meter, i.e., a three-dimensional spatial voxel set.

[0063] The second step involves performing probability mapping on each three-dimensional voxel in the aforementioned three-dimensional spatial voxel set, based on the probability distribution characteristics of the aforementioned fall scattering spatial model, to obtain a fall probability distribution value set. The aforementioned probability distribution characteristics can be a three-dimensional Gaussian distribution pattern formed in Monte Carlo simulations, where the fall probability is higher closer to the scattering center and lower closer to the edge. The aforementioned probability mapping process can be the integration and mapping of a continuous probability density function onto a discrete grid. The aforementioned fall probability distribution value set can be a set including the specific probability values ​​of each voxel being hit by the drone.

[0064] In practice, the three-dimensional probability density function of the fall scatter spatial model can be extracted, and the definite integral of this function over the spatial coordinate interval of each three-dimensional voxel in the three-dimensional voxel set can be calculated to obtain the absolute hit probability of the three-dimensional voxels, i.e., the fall probability distribution value. The fall scatter spatial model described above can utilize a three-dimensional anisotropic Gaussian distribution as the probability density function, i.e. .

[0065] Among them, the above It can be The above. This could be the crash site's dispersion center, the drone's malfunction point, or a predicted point in the wind field forecast. The above. It can be the covariance matrix. , Calibration can be achieved through Monte Carlo simulation (e.g., So, regarding voxels... The fall probability distribution value can be .

[0066] For example, we can determine that the fall probability distribution value of a certain three-dimensional spatial voxel located at the exact center of the scatter model is 0.05 (i.e., a 5% probability of being hit), while the value of the edge voxels is 0.001. This constitutes the set of fall probability distribution values.

[0067] The third step involves performing a correlation quantification calculation on each fall probability distribution value in the aforementioned fall probability distribution value set and the aforementioned mechanical impact kinetic energy to obtain a basic physical damage value set. This correlation quantification calculation can be achieved by introducing the concept of expected value from probability theory into the multiplicative operation process of physical damage assessment. The aforementioned basic physical damage value set can be the set of expected pure physical damage values ​​for each voxel, without considering the social attributes of the ground target.

[0068] In practice, each three-dimensional spatial voxel can be traversed, and the fall probability distribution value corresponding to the three-dimensional spatial voxel can be multiplied by the mechanical impact kinetic energy to obtain the energy expectation of the three-dimensional spatial voxel, that is, the basic physical damage value.

[0069] For example, for a voxel with a probability distribution value of 0.05, multiplying it by a mechanical impact kinetic energy of 4500 joules yields a basic physical damage value of 225 (joule probability), which is the basic physical damage value.

[0070] The fourth step involves applying social risk correction processing to each three-dimensional space within the aforementioned three-dimensional spatial voxel set, based on the aforementioned sensitivity level weights and the aforementioned basic physical damage value set, to obtain the incidental damage index. This social risk correction processing can be a process that introduces non-linear penalties related to the human and economic aspects of the ground target on top of pure physical damage.

[0071] In practice, the sensitivity level weight can be used as a parameter to comprehensively correct the basic physical damage value set through linear amplification and nonlinear bias functions, and then mapped onto a standardized risk management system. For example, although the basic physical damage value of a certain voxel is low, because it belongs to a chemical plant area (corresponding to a high weight), after social risk correction, the aforementioned incidental damage index derived from it jumps to the "red alert level".

[0072] In some optional implementations of certain embodiments, the execution entity may perform social risk correction processing on each three-dimensional space in the three-dimensional space voxel set based on the aforementioned sensitivity level weights and the aforementioned basic physical damage value set to obtain the incidental damage index, which may include the following steps: The first step involves parameter analysis of the aforementioned sensitivity level weights to obtain the exponential amplification factor and the risk baseline bias. The exponential amplification factor can be a scaling factor used to multiply and scale the baseline damage value to amplify the risk differences in high-risk areas. The risk baseline bias can be a constant additive factor used to ensure that even low-probability impact sensitive areas will generate a baseline alarm threshold.

[0073] In practice, multiplicative and additive components can be separated from a single sensitivity level weight value through table lookup or decoding algorithms. For example, from a sensitivity level weight value of "1.5", its exponential amplification factor is 1.2 and the risk baseline bias is 50.

[0074] The second step involves performing a product amplification operation on the corresponding basic physical damage values ​​of the aforementioned three-dimensional voxels based on the aforementioned exponential amplification factor, to obtain the initial corrected damage value. This product amplification operation can be a scalar multiplication mathematical calculation. The initial corrected damage value can be transitional risk data that has undergone scaling but has not yet had a baseline penalty term added.

[0075] In practice, the aforementioned execution entity can invoke the floating-point multiplier (FPU) in the processor to multiply the base physical damage value of each voxel by the exponential amplification factor. For example, multiplying the base physical damage value "225" by the exponential amplification factor "1.2" yields an initial corrected damage value of "270".

[0076] The third step involves superimposing the initial corrected damage value and the risk baseline bias to obtain the absolute risk value of the three-dimensional voxel. This superposition calculation can be a scalar addition operation. The absolute risk value can be the final risk quantification score of the voxel after considering physical probability and kinetic energy penalties.

[0077] In practice, an addition logic unit can be invoked to add the risk baseline bias to the initial corrected damage value. For example, adding the risk baseline bias of 50 to the initial corrected damage value of "270" yields an absolute risk value of "320" for the three-dimensional voxel.

[0078] The fourth step involves spatially integrating and summing the absolute risk values ​​of each three-dimensional voxel to obtain the global cumulative risk value. This spatial integration and summing process can be a numerical calculation method that comprehensively accumulates the values ​​in the discrete voxel grid. The global cumulative risk value can be a total score reflecting the overall secondary disaster threat level across the entire overlapping area.

[0079] In practice, a loop-based accumulation algorithm (e.g., accumulator register iteration) can be executed to sum the absolute risk values ​​of each 3D spatial voxel in the aforementioned 3D spatial voxel set. For example, by summing the absolute risk values ​​of 500 3D spatial voxels (e.g., 320, 310, 200, etc.), the aforementioned global cumulative risk value is obtained as "145000".

[0080] The fifth step involves performing a nonlinear normalized scaling transformation on the aforementioned global cumulative risk value to obtain a standard risk scalar. This nonlinear normalized scaling transformation can be a process that uses specific mathematical functions (e.g., the sigmoid function, logarithmic function, and extremum compression function) to compress or map the original data to a fixed standard interval (e.g., between 0 and 100). The standard risk scalar can be a standardized percentage score used for system comparisons.

[0081] In practice, the aforementioned global cumulative risk value can be substituted into a preset Sigmoid normalization formula. Through exponential and division operations, the influence of excessively large extreme values ​​can be suppressed, and the intermediate sensitive range can be stretched. For example, substituting the global cumulative risk value of 145000 into the normalization function yields a compressed mapping to "85.4", which is the standard risk scalar. The Sigmoid function can be... Among them, the above It can be a steepness parameter ( The above. It can be a center point parameter (e.g., The above. It could be the current global cumulative risk value (e.g., ).

[0082] Step 6: Based on a pre-defined risk control level table, determine the risk control level of the aforementioned standard risk scalars to obtain the collateral damage index. The pre-defined risk control level table can be a pre-configured data mapping dictionary in a database, where different standard score ranges correspond to different alarm levels (e.g., 0-20: low risk, 20-40: medium risk, 40-70: high risk, 70-100: extremely high risk). The aforementioned risk control level can be a risk classification represented by discrete labels (e.g., low, medium, high, extremely high).

[0083] In practice, a standard risk scalar can be used as a comparison input to execute interval threshold judgment logic (e.g., If-Else decision tree) in the aforementioned control risk level table to obtain the collateral damage index. For example, the standard risk scalar "85.4" can fall within the "80-100 (extremely high risk)" range, determining the control risk level as "extremely high (Level 4)," and using the combined data packet of "85.4 / Level 4" as the collateral damage index.

[0084] Step 105: Based on the communication link status and collateral damage index, generate graded countermeasure commands for the target UAV.

[0085] In some embodiments, the executing entity may generate tiered countermeasure commands against the target UAV based on the communication link status and the collateral damage index. These tiered countermeasure commands may be digital control messages generated by a computer system, possessing different intervention strategies (e.g., induced departure, forced landing, hard-kill crash) and underlying radio frequency parameters.

[0086] As an example, when the communication link status is identified as "unencrypted civilian image transmission protocol," if the collateral damage index of the target crashing on the spot is "90 (extremely high risk)," a graded countermeasure command is generated to "deceive navigation and guide it to a safe zone" to avoid the target crashing in densely populated areas. Conversely, if the collateral damage index is "10 (extremely low risk, such as in wasteland)," a graded countermeasure command is generated to "suppress the target across the entire frequency band at maximum power," forcing it to crash rapidly on the spot.

[0087] In some optional implementations of certain embodiments, after generating the graded countermeasure command against the target UAV based on the communication link status and the incidental damage index, the following steps may also be included: The first step, in response to the execution of the aforementioned tiered countermeasure command, is to collect multidimensional motion characteristic data of the target UAV. This multidimensional motion characteristic data can be a continuous physical sequence of spatial displacements described by the detection sensors during the period when the radio frequency countermeasure equipment transmits interference or deception signals to the target UAV, indicating whether the target is under control or out of control.

[0088] In practice, a high-frequency detection mission can be initiated at the same time as the command is issued. The target UAV's three-dimensional spatial coordinates, velocity, and attitude changes can be continuously acquired through radar or photoelectric tracking equipment at a set sampling frequency (e.g., 20 times per second).

[0089] For example, starting from the first second after the jamming signal is transmitted, the latitude, longitude and altitude of the target are collected every 50 milliseconds to form multidimensional motion feature data including position and velocity vectors.

[0090] The second step involves trajectory fitting of the aforementioned multidimensional motion feature data to obtain the real-time motion vector. This trajectory fitting can be a calculation process that uses the least squares method to derive the target's true motion path function from a set of discrete coordinate points with measurement noise. The real-time motion vector can be a three-dimensional physical vector reflecting the instantaneous flight direction and speed of the target UAV within the current extremely short time window.

[0091] In practice, multidimensional motion feature data from the most recent sliding window (e.g., the past 2 seconds) can be extracted and input into a curve fitting model to filter out abrupt noise caused by radar clutter. The first derivative of the fitted smooth curve can then be calculated to obtain the current instantaneous velocity and direction. For example, by performing least-squares fitting on 40 spatial coordinate points acquired in the past 2 seconds, the current real-time motion vector of the target UAV can be obtained as: 8 m / s horizontally southward and 3 m / s vertically downward.

[0092] The third step involves mapping the real-time motion vector of the target onto the aforementioned fall dispersion space model to determine the physical position offset of the target UAV relative to the expected landing area. The expected landing area can be the central fall region with the highest probability density in the fall dispersion space model, based on theoretical predictions. The physical position offset can be the three-dimensional Euclidean distance vector between the spatial position the target UAV is expected to reach based on its current actual flight trend and the position predicted by the theoretical model.

[0093] In practice, the aforementioned real-time motion vector can be used as initial momentum to perform forward extrapolation in the short-term spatial domain, calculate the predicted actual landing point, and then perform geometric subtraction between the coordinates of the actual landing point and the center coordinates of the expected landing area to obtain the distance difference between the two on the three spatial axes of X, Y, and Z.

[0094] For example, by short-term simulation, it was found that the actual landing point of the drone was affected by crosswinds and was biased to the east. Then, the distance difference between the actual landing point and the center of the expected landing area was calculated, resulting in a physical position offset of 15 meters in size and directed to the east.

[0095] The fourth step involves determining the impact of the aforementioned physical location offset on the collateral damage index based on the data from the sensitive area, thereby obtaining a real-time risk deviation value. This impact information can be the quantified risk fluctuation caused by deviations in the target's actual fall trajectory, resulting in it moving closer to or further from a ground hazard source (e.g., a gas station). The real-time risk deviation value can be the numerical difference between the actual risk level and the initially predicted collateral damage index.

[0096] In practice, the corrected landing point of the physical position offset can be re-substituted into the collision detection model to calculate its spatial distance from the aforementioned sensitive area data. If the distance is shortened, an increased risk score is calculated. If the distance is increased, a decreased risk score is calculated, which can be subtracted from the original risk score.

[0097] For example, calculations show that shifting 15 meters eastward brings the target closer to a school (i.e., sensitive area data), causing the risk score to rise from 60 to 75. The difference between the two (+15 points) can be used as the real-time risk deviation value.

[0098] Fifth, based on the aforementioned real-time risk deviation value, dynamic compensation and adjustment processing is performed on the aforementioned graded countermeasure commands to ensure that the actual fall path of the target UAV tends towards a preset safety envelope. This dynamic compensation and adjustment involves adjusting the interference intensity, transmission power, and guidance frequency in the graded countermeasure commands. This dynamic compensation and adjustment process can be a closed-loop control process that modifies the underlying radio frequency hardware operating parameters in real time based on the error feedback signal output by the system. The aforementioned safety envelope can be a virtual three-dimensional safe flight tunnel or isolation boundary pre-defined in three-dimensional space, ensuring that the target UAV falls within its range without touching any sensitive areas. The aforementioned interference intensity can be the degree to which the countermeasure signal suppresses the original communication signal-to-noise ratio of the target UAV. The aforementioned transmission power can be the energy of the electromagnetic waves radiated into space by the antenna of the countermeasure device. The aforementioned guidance frequency can be a specific radio frequency point used to induce the UAV to follow during navigation deception or link takeover.

[0099] In practice, the safety envelope can be used as the control target, and the real-time risk deviation value can be used as the error input of the PID controller. When the target UAV is in navigation deception or link takeover mode, the controller calculates the compensation coordinates or yaw commands required to correct the course, adjusts the transmission configuration of the software-defined radio (SDR) module through the internal bus, and dynamically modifies the GPS coordinate offset or joystick position carried in the deception signal or the remote control channel.

[0100] For example, when a real-time risk deviation of +15 points (increased risk) is detected and the target is biased towards the danger zone to the east, the SDR module is controlled through the digital interface to inject a virtual coordinate increment to the west into the generated navigation deception signal. This forces the UAV's flight control system to correct its course to the left (west) in order to correct the "false position" until its actual crash or landing path falls within the safety envelope.

[0101] In some optional implementations of certain embodiments, the execution entity may perform dynamic compensation and adjustment processing on the graded countermeasure instructions based on the real-time risk deviation value, which may include the following steps: The first step, based on the aforementioned real-time risk deviation value, is to extract the deviation vector and acceleration variation characteristics of the target UAV relative to the aforementioned safety envelope. The deviation vector can be the shortest normal distance vector from the target UAV's current actual spatial coordinates to the nearest boundary point of the safety envelope. The acceleration variation characteristics can be the rate of change matrix of the target UAV's velocity vector over time, which can be used to characterize the UAV's current maneuver intensity and resistance level. These acceleration variation characteristics can include linear acceleration characteristics and angular acceleration characteristics. The linear acceleration characteristic can be the rate of change of the target UAV's velocity vector over time. The angular acceleration characteristic can be the second derivative of the rate of change of the target UAV's flight attitude angles (pitch, roll, yaw) over time.

[0102] In practice, analytical geometry algorithms can be used to determine the perpendicular vector from the target's current coordinates to the safety envelope surface, which serves as the deviation vector. Then, the real-time motion vector is differentiated twice with respect to time to extract the linear acceleration characteristics, including normal and tangential components. By differentiating the flight attitude angle sequence in the aforementioned dynamic state parameters in the time dimension, or by differentiating the attitude angular velocity once, angular acceleration characteristics characterizing the severity of roll, pitch, or yaw are obtained.

[0103] For example, the deviation vector could be that the target is 5 meters away from the boundary of the safety envelope (direction outward). Calculations show its linear acceleration characteristic is 2 m / s², and its yaw acceleration is 30 rad / s². This indicates that the target UAV is in a highly maneuverable state, not only attempting to accelerate out of the safe zone in spatial displacement, but also undergoing drastic attitude adjustments to counter the current electromagnetic suppression, thus indicating a high degree of "resistance".

[0104] The second step involves matching the corresponding power gain coefficient and frequency step value within a preset compensation parameter matrix, based on the aforementioned deviation vector and acceleration variation characteristics. This preset compensation parameter matrix can be a two-dimensional or multi-dimensional look-up table pre-stored in memory, mapping radio compensation strategies under different deviation distances and acceleration states. The power gain coefficient can be a multiplier factor used to proportionally amplify or reduce the current transmit power. The frequency step value can be a specific frequency interval value used to fine-tune the offset of the current countermeasure operating frequency.

[0105] In practice, the scalar value of the aforementioned deviation vector can be used as the row index of the matrix, and the value of the aforementioned acceleration change characteristic can be used as the column index of the matrix. Cross-addressing can be performed in the aforementioned preset compensation parameter matrix to read the value in the corresponding storage unit. For example, using a deviation of 5 meters and an acceleration of 2 m / s² as indices, the corresponding operating parameters can be found in the compensation parameter matrix, and the power gain coefficient can be extracted as "1.5 times" and the frequency step value as "+500kHz".

[0106] The third step involves injecting and updating parameters in real time for the aforementioned tiered countermeasure instructions based on the power gain coefficient and frequency step value. This real-time parameter injection and update can be an overwrite operation that writes newly calculated underlying hardware configuration parameters into device registers or memory buffers to change the transient output state of the RF chip.

[0107] In practice, the current transmit power value can be multiplied by the power gain coefficient, and a frequency step value can be added to the current operating frequency. This repackages the data into a low-level communication protocol frame, which is then sent to the digital baseband control board of the jamming transmitter via a high-speed serial port or Ethernet port. For example, the current 10W transmit power can be multiplied by 1.5 to update it to 15W. The current 2.41GHz pilot frequency can be increased by 500kHz to update it to 2.4105GHz, and these new parameters can be injected into the RF transmit module in real time to take effect.

[0108] The fourth step involves determining the current compensation adjustment parameters and outputting control commands in response to the detection that the real-time motion vector of the target UAV and the aforementioned safety envelope meet the preset convergence conditions. The preset convergence conditions can be mathematical thresholds indicating that the target's motion trend has been completely corrected and there is no longer a risk of escape (e.g., the deviation vector magnitude is close to 0 for three consecutive cycles, and the acceleration change characteristics show a deceleration state). The compensation adjustment parameters can be the power and frequency configuration values ​​after iterative stabilization. The control commands can be execution commands requiring the underlying equipment to lock the current optimal parameters and maintain a constant suppression state.

[0109] In practice, the deviation vector can be continuously calculated by monitoring the feedback loop. When the magnitude of the deviation vector is found to be less than the tolerance threshold within a continuously set time period, a convergence interruption is triggered, freezing the current RF parameters and preventing further adjustments. For example, after continuous adjustment, if the target drone is detected to have completely returned to the safe envelope (deviation vector close to 0 meters) and lost power, beginning a vertical descent (meeting the preset convergence condition), the current 15W power and 2.4105GHz frequency are then confirmed as compensation adjustment parameters, and a control command to "lock parameters and execute constant suppression" is output to the hardware, completely eliminating the risk of collateral damage.

[0110] The various embodiments disclosed above have the following beneficial effects: The low-altitude UAV control method based on multimodal data fusion, as described in some embodiments of this disclosure, can achieve accurate probability prediction of the crash trajectory of out-of-control UAVs and quantitative assessment of the risk of secondary ground damage. This solves the ground safety hazards caused by the uncertainty of the crash during countermeasures, achieving a balance between control effectiveness and ground safety. Specifically, the high risk of secondary disasters related to low-altitude control and the blind nature of countermeasures are due to the fact that existing technologies often simplify UAVs to ideal linear crash points when implementing countermeasures, without considering the nonlinear dynamic failure evolution process of the target under the influence of complex meteorological wind fields. They cannot distinguish the attribute differences between densely populated areas (e.g., schools, hospitals) and low-risk areas (e.g., green spaces, water bodies), leading to blind countermeasure decisions. Based on this, the low-altitude UAV control method based on multimodal data fusion of some embodiments of this disclosure first, in response to a warning event detecting a target UAV in the target airspace, acquires multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace. This ensures that the system can grasp the electromagnetic and physical characteristics of the UAV before countermeasures, as well as identify the geographical constraints of the UAV's location (e.g., schools, hospitals), providing detailed data input for subsequent simulations. Secondly, the multi-source detection feature data is fused and analyzed to obtain the dynamic state parameters and communication link status of the target UAV. This enables deep perception of the target's real-time flight inertia and controllability. In real-world scenarios, it can distinguish whether the target is in a stable hovering or high-speed dive state, and assess the anti-interference strength of the remote control link, providing a physical basis for determining the countermeasure entry point. Then, based on the dynamic state parameters and wind field data from the environmental geographic feature data, a spatial model of the target UAV's fall dispersion under signal interference conditions is constructed. This transforms traditional single-landing-point prediction into a three-dimensional spatial envelope with probability distribution characteristics. In actual urban low-altitude scenarios, considering wind fields between high-rise buildings (e.g., through drafts) and the UAV's roll after loss of control, the dynamic drift path of the UAV after interference can be reconstructed, reducing prediction bias and improving the robustness of prediction results in complex weather conditions. Furthermore, based on the aforementioned crash dispersion spatial model, collision detection processing is performed on the sensitive area data in the aforementioned environmental geographic feature data to obtain a collateral damage index. Therefore, by spatially overlaying the predicted crash probability distribution with sensitive ground targets (e.g., schools, hospitals, and oil depots), the potential social impact and material (e.g., building materials) loss induced by countermeasures can be assessed in advance. Finally, based on the aforementioned communication link status and the aforementioned collateral damage index, tiered countermeasure commands are generated for the aforementioned target UAVs. This enables on-demand output of control commands. In actual countermeasures, if the predicted impact area is determined to have an extremely high risk of damage, the system can adopt flexible measures such as reducing interference power and guiding relocation.If the risk is within a controllable range, a forced landing is executed. This avoids administrative collateral damage caused by interference from a single execution method. Furthermore, this embodiment, through deep decoupling and reconstruction of multimodal data, digitally monitors the out-of-control evolution process, improving the scientific nature of decision-making and the safety of execution in handling sudden drone intrusion incidents by the low-altitude security system, and providing reliable technical support for the routine management and control of drones in complex urban environments.

[0111] Based on the same inventive concept, embodiments of this application provide a low-altitude unmanned aerial vehicle (UAV) control device based on multimodal data fusion, comprising: The data acquisition module is used to collect multi-source detection feature data and environmental geographic feature data; A memory for storing a program for a low-altitude unmanned aerial vehicle (UAV) management method based on multimodal data fusion; The processor and memory can load and execute programs to implement a low-altitude UAV management method based on multimodal data fusion.

[0112] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0113] This application provides a computer-readable storage medium storing a computer program that can be loaded and executed by a processor, which is a method for controlling low-altitude unmanned aerial vehicles based on multimodal data fusion.

[0114] Computer storage media include, for example, USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media that can store program code.

[0115] Based on the same inventive concept, this application provides a smart terminal, including a memory and a processor. The memory stores a computer program that can be loaded and executed by the processor, which is a low-altitude unmanned aerial vehicle (UAV) control method based on multimodal data fusion.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0117] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion, characterized in that, include: In response to an early warning event that a target UAV is detected in the target airspace, the system acquires multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace. The multi-source detection feature data is fused and analyzed to obtain the dynamic state parameters and communication link status of the target UAV; Based on the dynamic state parameters and the wind field data in the environmental geographic feature data, a crash dispersion spatial model of the target UAV under signal interference conditions is constructed. Based on the aforementioned fall dispersion spatial model, collision detection processing is performed on the sensitive area data in the environmental geographic feature data to obtain the incidental damage index. Based on the communication link status and the collateral damage index, a tiered countermeasure command is generated for the target UAV. 2.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 1, wherein, After generating graded countermeasure commands against the target UAV based on the communication link status and the collateral damage index, the method further includes: In response to the execution of the graded countermeasure command, multi-dimensional motion feature data of the target UAV are collected; Trajectory fitting is performed on the multidimensional motion feature data to obtain real-time motion vectors; The real-time motion vector of the target is mapped to the fall dispersion space model to determine the physical position offset of the target UAV relative to the expected landing area; Based on the sensitive area data, the impact of the physical location offset on the incidental damage index is determined, and a real-time risk deviation value is obtained. Based on the real-time risk deviation value, the graded countermeasure command is dynamically compensated and adjusted so that the actual crash path of the target UAV tends to the preset safety envelope. The dynamic compensation and adjustment is to adjust the interference intensity, transmission power and guidance frequency in the graded countermeasure command. 3.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 2, characterized in that, The step of dynamically compensating and adjusting the graded countermeasure command based on the real-time risk deviation value includes: Based on the real-time risk deviation value, extract the deviation vector and acceleration change characteristics of the target UAV relative to the safety envelope; Based on the deviation vector and the acceleration change characteristics, the corresponding power gain coefficient and frequency step value are matched in the preset compensation parameter matrix; Based on the power gain coefficient and the frequency step value, the parameters of the graded countermeasure command are injected and updated in real time. In response to the detection that the real-time motion vector of the target UAV and the safety envelope meet the preset convergence conditions, the current compensation adjustment parameters are determined and control commands are output. 4.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 1, characterized in that, The step of acquiring multi-source detection feature data of the target UAV and environmental geographic feature data of the target airspace in response to a warning event that a target UAV is detected in the target airspace includes: By calling upon a heterogeneous array of detection sensors deployed around the target airspace, the electromagnetic scattering characteristics, radio frequency communication characteristics, and photoelectric imaging characteristics of the target UAV are collected in parallel to obtain multi-source detection characteristic data. Based on a preset geographic information system, a digital elevation model, three-dimensional ground feature vector data, and a preset sensitive target distribution map that match the spatial coordinates in the multi-source detection feature data are determined. Based on meteorological monitoring nodes, wind speed, wind direction, and air density vectors at corresponding altitude levels are collected within the target airspace. The digital elevation model, the three-dimensional ground feature vector data, the preset sensitive target distribution map, the wind speed information, the wind direction information, and the air density vector are spatially superimposed to obtain environmental geographic feature data. 5.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 1, wherein, The step of constructing a spatial model of the target UAV's crash dispersion under signal interference conditions based on the dynamic state parameters and wind field data in the environmental geographic feature data includes: The dynamic state parameters are analyzed to obtain the initial spatial coordinates, initial motion vector, and flight attitude information of the target UAV. Based on the preset signal interference failure mode, the initial spatial coordinates, the initial motion vector and the flight attitude information are subjected to failure dynamics deduction to obtain the reference fall trajectory equation; Determine the wind field data in the environmental geographic feature data; Based on the wind field data, the baseline fall trajectory equation is subjected to multi-step Monte Carlo perturbation simulation to obtain a set of probabilistic fall trajectories; A three-dimensional spatial envelope fitting process is performed on each probability fall trajectory in the set of probability fall trajectories to obtain a fall scattering spatial model. 6.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 1, characterized in that, The step of performing collision detection processing on sensitive area data in the environmental geographic feature data based on the fall dispersion spatial model to obtain the incidental damage index includes: Vectorized boundary extraction processing is performed on the sensitive area data in the environmental geographic feature data to obtain the spatial outline of the sensitive area; A three-dimensional spatial intersection calculation is performed on the aforementioned fall dispersion spatial model and the spatial contour of the sensitive area to obtain the overlap range of the target UAV's landing area; The dynamic state parameters within the overlapping area of ​​the impact zone are processed by kinetic energy conversion to obtain the mechanical impact kinetic energy at the moment of the expected collision. The geographic element attributes within the overlapping area are subjected to level mapping processing to obtain the sensitivity level weight; Based on the overlapping range of the impact area, the mechanical impact kinetic energy and the sensitivity level weight are subjected to damage quantification processing to obtain the incidental damage index.

7. The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion according to claim 6, characterized in that, The step of quantifying the mechanical impact kinetic energy and the sensitivity level weight based on the overlap range of the impact area to obtain the incidental damage index includes: The overlapping range of the landing area is spatially discretized to obtain a three-dimensional spatial voxel set; Based on the probability distribution characteristics of the fall scattering spatial model, probability mapping is performed on each three-dimensional spatial voxel in the three-dimensional spatial voxel set to obtain a fall probability distribution value set. For each fall probability distribution value in the fall probability distribution value set, the fall probability distribution value and the mechanical impact kinetic energy are correlated and quantitatively calculated to obtain a basic physical damage value set. Based on the sensitivity level weights and the basic physical damage value set, social risk correction processing is performed on each three-dimensional space in the three-dimensional spatial voxel set to obtain the incidental damage index. 8.The low-altitude unmanned aerial vehicle management and control method based on multi-modal data fusion of claim 7, characterized in that, The step of performing social risk correction processing on each three-dimensional space in the three-dimensional space voxel set based on the sensitivity level weight and the basic physical damage value set to obtain the incidental damage index includes: The sensitivity level weights are analyzed to obtain the exponential amplification factor and the risk baseline bias. Based on the exponential amplification factor, the corresponding basic physical damage value of the three-dimensional spatial voxel is subjected to a product amplification operation to obtain the initial corrected damage value. The initial corrected damage value and the risk baseline bias are superimposed and calculated to obtain the absolute risk value of the three-dimensional spatial voxel; The global cumulative risk value is obtained by summing the absolute risk values ​​of each three-dimensional spatial voxel through spatial integration. The global cumulative risk value is subjected to nonlinear normalization scaling transformation to obtain a standard risk scalar; Based on a preset risk level table, the risk level of the standard risk scalar is determined, and the collateral damage index is obtained.

9. A low-altitude unmanned aerial vehicle (UAV) control device based on multimodal data fusion, characterized in that, include: The data acquisition module is used to collect multi-source detection feature data and environmental geographic feature data; A memory for storing a program of a low-altitude unmanned aerial vehicle (UAV) control method based on multimodal data fusion as described in any one of claims 1 to 8; The processor and the program in the memory can be loaded and executed by the processor to implement the low-altitude UAV control method based on multimodal data fusion as described in any one of claims 1 to 8.

Images