Unmanned aerial vehicle automatic capturing system based on artificial intelligence

Abstract

The invention relates to the technical field of unmanned aerial vehicles, and discloses an unmanned aerial vehicle automatic capture system based on artificial intelligence, and the system comprises a dynamic environment sensing module and a capture decision module. The multi-modal spatial data of the target unmanned aerial vehicle are collected in real time, and the multi-modal spatial data comprise three-dimensional coordinates (fusing data of a multispectral radar and a laser array sensor), flight attitudes and electromagnetic characteristic data; after dynamic interference suppression is carried out on the data, processing is carried out through a decision processing layer (including a data cleaning module and a collaborative modeling module); the data cleaning module completes track segmentation, noise filtering and standardized calibration; the collaborative modeling module is trained through multi-scene historical data, an electromagnetic interference analysis layer of the collaborative modeling module generates electromagnetic coupling characteristic data, a trajectory correction layer models trajectory association and generates compensation trajectory field data, and a decision execution layer fuses the data and outputs a dynamic interception instruction. According to the system, through multi-modal fusion and dynamic modeling, the accuracy and adaptability of unmanned aerial vehicle capturing in a complex environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of drones, and in particular to an automatic drone capture system based on artificial intelligence. Background Art

[0002] With the rapid development of drone technology, drones are increasingly being used in civil, industrial, and military applications. However, their widespread use also raises a series of security and management issues. For example, "illegal" drones that illegally intrude into designated airspace can threaten public safety, leak sensitive information, and even be used for malicious attacks. Traditional drone control methods, such as manual visual tracking and radio jamming, suffer from limitations such as slow response, low accuracy, and significant environmental impact. These limitations make them inadequate for efficient and precise drone capture in complex and dynamic environments.

[0003] Existing drone capture technologies, based on single-sensor monitoring and control methods, generally suffer from a single data dimension. For example, relying solely on radar for positioning can lead to large positioning errors due to electromagnetic interference; relying solely on visual sensors can significantly degrade performance in complex weather or low-light conditions. Furthermore, traditional decision-making and control systems often utilize fixed algorithm models that lack the ability to adapt to dynamic environments. This inability to flexibly adjust strategies to varying interception scenarios (such as urban complexes and open fields) results in inconsistent capture success rates.

[0004] Furthermore, the application of multimodal data fusion technology in drone control still faces numerous challenges. Data collected by different sensor types (such as radar, laser array sensors, and electromagnetic signature acquisition devices) exhibit varying temporal and spatial characteristics. Efficient cleaning, feature extraction, and dynamic correlation modeling of multi-source data are crucial for improving system accuracy and reliability. Existing technologies often employ fixed-threshold filtering in the data cleaning process, making it difficult to effectively handle dynamically changing noisy data. Furthermore, trajectory modeling and interference analysis lack the ability to combine electromagnetic signatures with spatial trajectory analysis, leading to biased judgments about the drone's true intentions.

[0005] When it comes to trajectory correction and interception command generation, traditional systems typically intercept based on a preset fixed path, failing to fully consider the drone's real-time flight attitude and electromagnetic signature changes. For example, when a target drone performs evasive maneuvers, traditional systems are unable to quickly adjust the interception path, easily leading to capture failure due to trajectory deviation. Furthermore, electromagnetic interference and trajectory control lack synergy, failing to form a complete closed-loop feedback mechanism, resulting in weak anti-interference capabilities in complex electromagnetic environments. Summary of the Invention

[0006] The purpose of the present invention is to provide an artificial intelligence-based drone automatic capture system to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based drone automatic capture system, the system comprising: A dynamic environment perception module is used to acquire multimodal spatial data of the target UAV in real time. The multimodal spatial data includes a first monitoring sequence corresponding to three-dimensional coordinate data, a second monitoring sequence corresponding to flight attitude data, and a third monitoring sequence corresponding to electromagnetic signature data. The three-dimensional coordinate data includes first spatial positioning data generated by a multispectral radar and second dynamic trajectory data collected by a laser array sensor. A capture decision module is used to perform dynamic interference suppression processing on the multimodal spatial data and input the data into a decision processing layer for feature modeling, and generate dynamic interception instructions according to the output results of the decision processing layer; The decision processing layer includes a data cleaning module and a collaborative modeling module, wherein the data cleaning module is used to segment the trajectory of the original spatial data stream and filter the noise data. The collaborative modeling module is obtained by joint training based on the historical trajectory data and historical electromagnetic data of multiple interception scenarios; the collaborative modeling module includes an electromagnetic interference analysis layer, a trajectory correction layer and a decision execution layer connected in sequence.

[0008] Preferably, the electromagnetic interference analysis layer is used to perform time domain correlation processing on different monitoring sequences in the original spatial data stream to generate electromagnetic coupling feature data; the trajectory correction layer is used to model the dynamic trajectory correlation relationship between the electromagnetic coupling feature data corresponding to each monitoring sequence to generate compensated trajectory field data; the decision execution layer is used to perform multi-dimensional fusion based on the compensated trajectory field data and the electromagnetic coupling feature data to generate dynamic interception instructions.

[0009] Preferably, the modeling of the dynamic trajectory correlation relationship between the electromagnetic coupling characteristic data corresponding to each monitoring sequence to generate the compensation trajectory field data includes: Using a trajectory optimization algorithm to identify key flight nodes in the electromagnetic coupling characteristic data, and determining a trajectory compensation sequence corresponding to each monitoring sequence based on the airspace type corresponding to each key flight node; Calculate the offset coefficient between the flight nodes in the same airspace in the trajectory compensation sequence corresponding to any two monitoring sequences, and generate the compensation trajectory field data between the any two monitoring sequences based on the offset coefficient.

[0010] Preferably, the calculation of the offset coefficient between flight nodes in the same airspace in the trajectory compensation sequences corresponding to any two monitoring sequences includes: When the number of flight nodes in the trajectory compensation sequence corresponding to any two monitoring sequences is inconsistent, virtual trajectory point interpolation is performed based on the airspace parameters corresponding to the terminal flight node in the one with fewer flight nodes, and the offset coefficient between the flight nodes in the same airspace is calculated based on the interpolated data.

[0011] Preferably, the data cleaning module is specifically used to: Dividing the first monitoring sequence, the second monitoring sequence, and the third monitoring sequence into equal density according to a preset spatial interval to generate standardized first spatial data, standardized second posture data, and standardized third electromagnetic data; A dynamic path clustering method is used to perform real-time calibration on the standardized first spatial data and the standardized second posture data, and a fixed spectrum analysis method is used to perform steady-state optimization on the standardized third electromagnetic data, and a first calibration sequence, a second calibration sequence, and a third calibration sequence are output; wherein the first calibration sequence includes the calibrated first spatial positioning data and the calibrated second dynamic trajectory data.

[0012] Preferably, the data cleaning module is further used to: Calculating a trajectory offset coefficient between the calibrated first spatial positioning data and the calibrated second dynamic trajectory data within a historical interception period; Predicting an expected trajectory value of the calibrated second dynamic trajectory data in the real-time interception period according to the trajectory deviation coefficient and the spatial parameters of the calibrated first spatial positioning data in the real-time interception period; Target interception compensation data is generated based on the calibrated second dynamic trajectory data and its expected trajectory value, and a monitoring sequence corresponding to the target interception compensation data is used as a first calibration sequence.

[0013] Preferably, the trajectory correction layer specifically includes: a path tracing unit, configured to perform trajectory path tracing for each monitoring sequence in the electromagnetic coupling characteristic data, so as to extract a corresponding trajectory propagation chain from each monitoring sequence; The spatial domain matching unit is used to perform spatiotemporal mapping between the trajectory propagation chain extracted from each monitoring sequence and the corresponding electromagnetic coupling feature data to generate compensated trajectory field data.

[0014] Preferably, the trajectory correction layer further includes: The signal anti-interference unit is used to perform electromagnetic hysteresis effect elimination processing on the compensation trajectory field data.

[0015] Preferably, the decision execution layer specifically includes: A multi-dimensional collaboration unit, comprising a plurality of interception collaboration nodes, each of which is connected to each monitoring sequence in the compensation trajectory field data and electromagnetic coupling characteristic data through permission configuration; a dynamic path optimization unit, configured to iteratively adjust the permission configuration using a dynamic path optimization algorithm to minimize the error between the dynamic interception instruction and the actual trajectory distribution; The trajectory anomaly recognition unit is used to locate the airspace threat area based on the compensated trajectory field data and the electromagnetic coupling characteristic data, and generate a dynamic interception instruction.

[0016] Preferably, the dynamic path clustering method specifically includes: Generate adaptive calibration parameters based on obstacle distribution characteristics of the real-time airspace environment; A spectrum analysis window mechanism is used to perform segmented calibration processing on the standardized first spatial data.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) In terms of dynamic environment perception, the system uses multispectral radar and laser array sensors to obtain the three-dimensional coordinate data (including the first spatial positioning data and the second dynamic trajectory data), flight attitude data and electromagnetic characteristic data of the target UAV, forming multimodal spatial data. The multi-dimensional data acquisition method overcomes the limitations of a single sensor. For example, the multispectral radar can maintain stable positioning capabilities under complex weather conditions, the laser array sensor can accurately capture the dynamic trajectory details of the UAV, and the electromagnetic characteristic data can reflect the working status and communication characteristics of the UAV. The combination of the three enables the system to fully and accurately perceive the real-time status of the target UAV, providing a rich information basis for subsequent decision-making; (2) The capture decision module generates dynamic interception instructions through dynamic interference suppression processing and feature modeling of the decision processing layer. The collaborative modeling module of the decision processing layer is jointly trained based on historical data of multiple interception scenarios. It can learn the flight patterns and electromagnetic characteristic patterns of drones in different scenarios, so that the system has the ability to adapt to different environments. For example, in urban building complexes, the system can identify common avoidance paths and electromagnetic interference patterns of drones based on historical data and adjust the interception strategy in advance; in open areas, it can use more efficient straight-line paths for interception, which significantly improves the decision-making efficiency and accuracy of the system in complex scenarios. (3) The data cleaning module standardizes and calibrates the raw data in real time through methods such as equal-density division of preset spatial intervals, dynamic path clustering, and fixed spectrum analysis. For example, real-time calibration of spatial data and attitude data using dynamic path clustering can effectively eliminate trajectory deviations caused by sensor errors or environmental interference; steady-state optimization of electromagnetic data can filter out background noise and extract real and effective electromagnetic features. The accuracy of the cleaned data is significantly improved, providing a reliable data basis for subsequent feature modeling and trajectory correction, and avoiding decision-making errors caused by raw data noise; (4) The trajectory correction layer realizes the spatiotemporal correlation modeling and anti-interference processing of electromagnetic coupling feature data and trajectory data through the path tracking unit, spatial matching unit and signal anti-interference unit. For example, the trajectory propagation chain extracted by the path tracking unit can clearly reflect the evolution of the UAV's flight path. Combined with the electromagnetic coupling feature data for spatiotemporal mapping, it can accurately identify the UAV's true flight intention and potential threat area; the signal anti-interference unit eliminates the electromagnetic lag effect, ensures the real-time and accuracy of the trajectory compensation data, enables the system to respond to the maneuver changes of the target UAV in a timely manner, and improves the interception success rate; (5) The multi-dimensional collaborative unit of the decision-making execution layer realizes the flexible scheduling and collaborative processing of multi-source data through authority configuration. The dynamic path optimization unit minimizes the interception error by iteratively adjusting the authority configuration. The trajectory anomaly identification unit accurately locates the airspace threat area based on multi-dimensional data. For example, the dynamic path optimization algorithm can continuously adjust the interception path according to real-time data to minimize the error between the dynamic interception instruction and the actual trajectory distribution; the trajectory anomaly identification unit can provide early warning of abnormal behavior of the target drone, gaining more reaction time for the system. The collaborative work of each unit forms a complete closed-loop control system, ensuring that the system can quickly and accurately generate the optimal interception strategy in complex environments; (6) In addition, when dealing with inconsistent numbers of monitoring sequences, the system achieves data alignment through virtual trajectory point interpolation, ensuring the accuracy and integrity of multi-source data fusion. This data processing mechanism effectively solves the problems caused by differences in sampling frequencies and data formats of different sensors, and enhances the system's compatibility and processing capabilities for heterogeneous data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a diagram showing the working principle of the artificial intelligence-based drone automatic capture system described in the present invention.

[0019] Figure 2 This is a diagram showing the working principle of electromagnetic interference analysis and trajectory correction in the decision-making processing layer.

[0020] Figure 3 Flowchart for modeling dynamic trajectory associations and generating compensated trajectory field data.

[0021] Figure 4 Flowchart of data standardization and calibration processing in the data cleaning module.

[0022] Figure 5 This is a diagram showing the working principle of trajectory path tracing and time-space mapping in the trajectory correction layer. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figure 1-Figure 5 The present invention relates to an artificial intelligence-based drone automatic capture system, which includes a dynamic environment perception module and a capture decision module. The specific implementation steps are as follows: The dynamic environment perception module is used to acquire multimodal spatial data of the target drone in real time. This multimodal spatial data includes a first monitoring sequence corresponding to three-dimensional coordinate data, a second monitoring sequence corresponding to flight attitude data, and a third monitoring sequence corresponding to electromagnetic signature data. The three-dimensional coordinate data includes first spatial positioning data generated by a multispectral radar and second dynamic trajectory data collected by a laser array sensor. The capture decision module is used to perform dynamic interference suppression processing on this multimodal spatial data and input it into the decision processing layer for feature modeling. Dynamic interception instructions are generated based on the output of the decision processing layer. The decision processing layer includes a data cleaning module and a collaborative modeling module. The data cleaning module is used to segment the trajectory of the raw spatial data stream and filter noise data. The collaborative modeling module is jointly trained based on historical trajectory data and historical electromagnetic data from multiple interception scenarios. The collaborative modeling module comprises an electromagnetic interference analysis layer, a trajectory correction layer, and a decision execution layer, which are connected in sequence.

[0025] Example 1:

[0026] The system's dynamic environment perception module collects multimodal spatial data using multiple sensor types. A multispectral radar continuously emits electromagnetic waves encompassing multiple frequency bands, including ultraviolet, visible, and infrared, to perform a three-dimensional scan of the target drone's location. The radar receiver analyzes the echo signal's frequency offset, phase shift, and intensity attenuation to generate real-time three-dimensional coordinate data for the target drone. This data includes X / Y / Z coordinates, flight velocity vector, and acceleration parameters, forming a time-indexed first monitoring sequence. The laser array sensor utilizes a distributed array structure, comprising multiple laser emitting and receiving units arranged in a matrix with preset angles, forming a laser monitoring network covering a specific airspace. When a target drone enters the monitoring area, a laser beam is projected onto the target surface and reflected back to the receiving unit. Using triangulation, the target's attitude parameters, including the real-time values and rates of change of roll, pitch, and yaw angles, are calculated. This generates a second dynamic trajectory data, forming a time-ordered second monitoring sequence. The electromagnetic signature acquisition unit integrates a broadband antenna and spectrum analyzer to monitor electromagnetic signals radiated by the target drone's communication links, power systems, and other components in real time. It extracts characteristic parameters such as the signal's center frequency, power spectrum density, modulation format, and pulse repetition frequency, forming a third monitoring sequence. All three types of data acquisition equipment are connected to a unified clock synchronization system. Using GPS timing or internal crystal oscillator synchronization, each set of monitoring data is timestamped with nanosecond accuracy, ensuring strict temporal alignment of multimodal data.

[0027] The input of the capture decision module is connected to the output of the dynamic environment perception module via a high-speed data bus. After receiving multimodal spatial data containing timestamps, dynamic interference suppression is first performed. During this process, the adaptive filtering unit, based on the minimum mean square error (LMS) algorithm, estimates the noise characteristics of the input data in real time and generates corresponding filter coefficients. This performs dual frequency and time domain suppression on pulse interference, Gaussian white noise, and narrowband interference in the data. For example, co-frequency interference generated by other drone communications in the environment is attenuated in the corresponding frequency band using a notch filter; periodic noise caused by mechanical vibration is dynamically tracked and eliminated using an adaptive notch filter. The data stream after interference suppression is delivered to the input buffer queue of the decision processing layer.

[0028] The data cleaning module reads the raw spatial data stream from the input buffer queue and first performs trajectory segmentation. The system presets a time window threshold (e.g., 500ms) or a spatial distance threshold (e.g., 10 meters). When the time interval between consecutive data points in the data stream exceeds the time threshold or the spatial displacement exceeds the distance threshold, segmentation is automatically performed, dividing the continuous data stream into multiple independent trajectory segments. Each trajectory segment contains a set of continuous first monitoring sequence, second monitoring sequence, and third monitoring sequence data, facilitating subsequent targeted processing of data from different flight phases (e.g., takeoff, cruise, and landing). In the noise data filtering stage, a cascade of median filtering and Kalman filtering is used: the median filter first sorts the neighborhood of each data point in each monitoring sequence, replaces outliers with median values, and eliminates salt and pepper noise; then, the Kalman filter predicts and updates the motion state of the target drone based on the state-space model, further smoothing the data curve and correcting measurement errors caused by random noise. For example, for the three-dimensional coordinate data in the first monitoring sequence, the Kalman filter recursively estimates the true state of the target by establishing the target's motion equation (such as a uniform motion model or a uniformly accelerated motion model) and combining it with the radar measurement value, and outputs a filtered coordinate sequence.

[0029] The training data storage unit of the collaborative modeling module pre-stores a massive amount of historical interception scenario data, covering a variety of terrain conditions such as open plains, urban buildings, mountain canyons, and different meteorological environments such as daytime, nighttime, rain and fog. Each set of historical data contains a complete multimodal monitoring sequence and the corresponding manually labeled interception strategy (such as interception path, interception timing, and interception method). During the joint training process, a deep learning framework is used to construct a multi-layer neural network model. The input layer receives normalized historical trajectory data and historical electromagnetic data, and the output layer corresponds to different interception strategy categories. The electromagnetic interference analysis layer, trajectory correction layer, and decision execution layer serve as the intermediate layers of the neural network. The weight parameters of each layer are optimized through the backpropagation algorithm to establish a nonlinear mapping relationship between multimodal data features and interception strategies. After the training is completed, the collaborative modeling module can perform feature extraction and pattern recognition on the input multimodal data in real time, analyze the time correlation characteristics between the data through the electromagnetic interference analysis layer in turn, optimize the trajectory prediction model through the trajectory correction layer, and finally generate a dynamic interception instruction containing interception path coordinates, speed instructions, attitude adjustment parameters and other information by the decision execution layer. The instruction is transmitted to the UAV capture execution agency (such as net capture device, electromagnetic jammer, etc.) through the data interface to realize automatic interception of the target UAV.

[0030] Example 2:

[0031] In the electromagnetic interference analysis layer of the collaborative modeling module, the temporal correlation processing of the raw spatial data stream is based on a spatiotemporal alignment mechanism for multimodal data. The system first synchronizes and aligns the first monitoring sequence (3D coordinate data), the second monitoring sequence (flight attitude data), and the third monitoring sequence (electromagnetic signature data) output by the dynamic environment perception module according to their timestamps, ensuring strict temporal correspondence between the three types of data. For example, within a specific sampling period (e.g., 100ms), the system extracts the 3D coordinate sequence (including position, velocity, and acceleration), the flight attitude sequence (roll, pitch, yaw angle, and their rates of change), and the electromagnetic signature sequence (frequency, power, and modulation) within that period to form a structured time series dataset. By calculating the cross-correlation function between the different sequences, the temporal correlation of the data is analyzed. For example, the time delay between a velocity mutation point in the 3D coordinates and a power spectrum peak in the electromagnetic signature, or the synchronization between flight attitude changes and electromagnetic signal frequency jumps, is calculated. This allows the identification of electromagnetic coupling signatures reflecting the target drone's behavior, such as multimodal data correlation patterns corresponding to specific flight maneuvers.

[0032] The trajectory correction layer models dynamic trajectory relationships as follows: First, a trajectory optimization algorithm (such as a modified version of the A* algorithm) is used to analyze electromagnetic coupling feature data. By setting feature change thresholds (e.g., coordinate change > 2 meters, attitude angle change > 15°, and electromagnetic frequency offset > 5 MHz), critical flight nodes, such as takeoff, landing, turning apex, and velocity extremes, are identified. Each critical flight node corresponds to a specific airspace type, determined by integrating environmental perception data from multispectral radar and laser array sensors. For example, multispectral radar object classification results are used to identify no-fly zone boundaries, while laser point cloud data is used to construct an obstacle distribution model to delineate dense and open areas. Based on different airspace types, the system generates differentiated trajectory compensation sequences for each monitoring sequence. At critical nodes at the edge of no-fly zones, the trajectory compensation sequence is forced to insert the highest intercept priority pathpoints to limit the target drone's approach. At critical nodes in open airspace, the trajectory compensation sequence is allowed to retain a certain adjustment margin to optimize interception energy consumption.

[0033] When calculating the offset coefficients of flight nodes in the same airspace within the trajectory compensation sequences of any two monitoring sequences (e.g., the first and third monitoring sequences), if the number of nodes differs, the system first interpolates virtual trajectory points for the sequence with fewer nodes. This interpolation rule is based on the motion parameters of the terminal node. For example, if the terminal node of the third monitoring sequence corresponds to a stable electromagnetic signature period (no frequency jumps, power fluctuations <3dB), a virtual node is generated based on the timestamp of that node and the average time interval between the preceding nodes (e.g., 200ms). The electromagnetic signature parameters of the virtual node retain the values of the terminal node. If the terminal node is in a signature change period (e.g., a linear frequency increase), linear extrapolation is used to generate the signature parameters of the virtual node. After interpolation, the nodes of the two sequences are mapped to the same spatial grid (e.g., a 5-meter cube grid) using a spatial region partitioning algorithm (e.g., octree partitioning). A pairwise matching of nodes within each grid is performed. For matching nodes, the Euclidean distance offset in the three-dimensional coordinate dimension, the Euler angle difference in the flight attitude dimension, and the cosine similarity distance in the electromagnetic feature dimension are calculated. The standardized offset coefficients are obtained through normalization processing (such as dividing by the maximum possible value of each dimension). Then, a weight is assigned to each dimension according to the airspace type (such as no-fly zone and ordinary zone) (for example, the coordinate offset weight in the no-fly zone accounts for 70%, and the electromagnetic feature accounts for 30%). After weighted summation, the compensated trajectory field data between the two monitoring sequences is generated. This data represents the degree of difference in the spatial and feature dimensions of the multimodal data, providing a quantitative basis for subsequent trajectory correction.

[0034] Example 3:

[0035] The data cleaning module begins processing the raw spatial data stream by dividing it into pre-defined airspace intervals. The system predefines these intervals based on specific application scenarios (such as urban low-altitude airspace control and campus security). For example, the 0-100-meter range is divided into a low-altitude monitoring zone, 100-500 meters into a medium-altitude monitoring zone, and 500 meters and above into a high-altitude monitoring zone. Each interval is then divided into equal-density zones. For the low-altitude monitoring zone, the system divides the space into grid cells of 10 meters in vertical height and 50 meters by 50 meters in horizontal plane. The first monitoring sequence (3D coordinate data), second monitoring sequence (flight attitude data), and third monitoring sequence (electromagnetic signature data) falling within each grid cell are resampled at equal intervals to ensure a consistent number of data points within each grid cell, generating standardized first spatial data, second attitude data, and third electromagnetic data. For example, if a grid cell contains 15 3D coordinate points in the raw data, the system resamples it to a fixed number of 10 points using linear interpolation to even out the spatial distribution of the data.

[0036] For the standardized first spatial data and second posture data, a dynamic path clustering method generates adaptive calibration parameters based on real-time obstacle distribution characteristics. Multispectral radar and laser array sensors scan the monitoring area in real time, generating a point cloud map containing the location, shape, and height of obstacles. Based on this map, the system calculates obstacle density values (e.g., the number of obstacle points per cubic meter) for each airspace. In areas with high obstacle density (e.g., above buildings), the calibration parameters for spatial positioning data are weighted 70% and those for posture data are 30%. Areas with sparse obstacles (e.g., open squares) use an equal weighting (50% each). A spectral analysis windowing mechanism divides the standardized first spatial data into multiple time windows (e.g., each containing 50 data points). A Fourier transform is performed on the data within each window to identify low-frequency noise below 0.1 Hz (typically caused by sensor drift) and high-frequency noise above 10 Hz (typically caused by mechanical vibration). A bandpass filter is then applied to retain the significant frequency components (0.1-10 Hz), enabling segmented calibration of the spatial positioning data.

[0037] For standardized third-order electromagnetic data, a fixed spectrum analysis method pre-sets the frequency range of interest (e.g., the 2.4GHz-5.8GHz drone communication band) and analysis parameters (e.g., a 100kHz resolution bandwidth and a 1MHz video bandwidth). The data is then spectrally scanned to extract steady-state characteristics such as power spectral density and center frequency stability within that band. By comparing the real-time spectrum with a historical reference spectrum, transient interference signals that deviate from the reference by more than a preset threshold (e.g., ±3dB) are suppressed, and a smoothed third-order calibration sequence is output.

[0038] The data cleaning module further calculates the trajectory offset coefficient between the calibrated first spatial positioning data and the second dynamic trajectory data within the historical interception period. The historical interception period is defined as the time interval of the past N successful interception processes (for example, N=10). The system calculates the root mean square error (RMSE) of the coordinate data (x1, y1, z1) and the position estimate (x2, y2, z2) derived from the posture data at the same time point in each historical period. The formula is:

[0039] Where M is the number of sampling points within each cycle, and the RMSE value is used as the trajectory offset coefficient. During the real-time intercept cycle, the system uses a linear regression model (based on historical training results of the offset coefficient and spatial parameters) to predict the expected trajectory value of the second dynamic trajectory data based on the airspace parameters of the current first spatial positioning data (such as coordinates, velocity, and grid cell). For example, if the current airspace is densely populated with low-altitude obstacles, and historical data shows that the average trajectory offset coefficient in this area is 1.2 meters, a -1.2-meter compensation is added to the position estimate derived from the attitude data during the prediction. Finally, based on the difference between the calibrated second dynamic trajectory data and the expected trajectory value, target interception compensation data containing position compensation vectors and attitude adjustment parameters is generated. The monitoring sequence corresponding to this data is used as the first calibration sequence and input into the collaborative modeling unit of the capture decision module.

[0040] Example 4:

[0041] The path tracking unit in the trajectory correction layer performs independent trajectory tracing for each monitoring sequence in the electromagnetic coupling signature data. For the first monitoring sequence (three-dimensional coordinate data), the system traverses each data point in chronological order, recording the target drone's spatial position at each moment, and constructing a continuous three-dimensional flight path, such as a broken-line or curved trajectory from starting point A (x1, y1, z1) through intermediate point B (x2, y2, z2) to end point C (x3, y3, z3). For the second monitoring sequence (flight attitude data), the system analyzes the temporal trends of the roll, pitch, and yaw angles to identify the target drone's motion pattern. For example, attitude parameters remain stable during straight, horizontal flight, while roll and yaw angles exhibit regular changes during turns, and the pitch angle increases rapidly during dives. For the third monitoring sequence (electromagnetic signature data), the system tracks the evolution of electromagnetic signal parameters such as frequency and intensity. For example, when the target drone initiates a communication link, a carrier component of a specific frequency appears in the electromagnetic signal, and the electromagnetic radiation intensity increases significantly when the motor accelerates. Through the above tracing process, a trajectory propagation chain reflecting the continuous changes in target behavior is extracted from each monitoring sequence, such as the time series chain of "rapid rise of three-dimensional coordinates during takeoff phase - attitude data showing an increase in climb angle - gradual increase in electromagnetic signal strength".

[0042] The spatial matching unit maps the trajectory propagation chain of each monitoring sequence with the corresponding electromagnetic coupling characteristic data in the spatiotemporal dimensions. Specifically, for each key time point in the trajectory propagation chain (such as takeoff, turn, and hover), the corresponding spatial position (such as longitude, latitude, and altitude), flight attitude parameters (such as roll angle θ = 5°, pitch angle φ = -10°), and electromagnetic characteristic parameters (such as frequency f = 2.45 GHz and power P = 20 dBm) are extracted. A "time-space-attitude-electromagnetic" four-tuple mapping relationship is established. For example, if the target drone is at spatial position (x0, y0, z0) at time t0, with a horizontal flight attitude (θ = 0°, φ = 0°), and the electromagnetic signal exhibits continuous wave characteristics (f = 2.4 GHz, P = 15 dBm), these data are associated to form a mapping node. By traversing the entire trajectory propagation chain, compensated trajectory field data containing a large number of mapping nodes is generated. This data is indexed by the space-time grid and records the corresponding posture features and electromagnetic features in each grid unit, realizing the fusion expression of multimodal data.

[0043] The signal anti-interference unit compensates for electromagnetic lag effects in trajectory field data. This lag manifests as a time delay between the target drone's physical actions (such as turning and acceleration) and the corresponding electromagnetic signal changes. For example, changes in electromagnetic radiation intensity caused by changes in motor speed may lag behind the actual action by 0.1-0.5 seconds. To mitigate this delay, the system pre-establishes a lag time model for typical actions using measured data. For example, "the average electromagnetic lag time for a sharp turn is 0.3 seconds." During real-time processing, the signal anti-interference unit queries the lag time model based on the action type in the trajectory propagation chain (e.g., a turn determined by posture data). It then performs a time-advance correction on the electromagnetic coupling signature data. This time-advances the electromagnetic signal data's timestamp by the corresponding lag time, aligning the electromagnetic signature with the physical action. For example, if the target UAV is detected to be performing a turning action at time t1, and the corresponding lag time of the action is 0.2 seconds, the electromagnetic characteristic data at time t1 is associated with the physical action data from t1-0.2 seconds to ensure the consistency of the temporal, spatial and electromagnetic characteristics in the compensation trajectory field data, thereby improving the real-time and accuracy of trajectory correction.

[0044] Example 5:

[0045] The multi-dimensional coordination unit in the decision-making execution layer utilizes a distributed architecture, comprising multiple independent intercept coordination nodes. Each node is assigned specific processing privileges to access specific monitoring sequences within the compensation trajectory field data and electromagnetic coupling signature data. For example, node A is configured to receive only the first monitoring sequence (three-dimensional coordinate data) and the second monitoring sequence (flight attitude data), responsible for analyzing the spatial motion trajectory of the target drone; node B receives only the third monitoring sequence (electromagnetic signature data), focusing on identifying communication signal patterns and powertrain status. Privileges are configured through predefined data flow routing rules, such as classifying input data by data labels ("coordinate data," "attitude data," and "electromagnetic data"), ensuring that each node only accesses the monitoring sequences within its scope of privilege. This distributed processing model enables parallel analysis of multimodal data, shortening the decision-making process.

[0046] The dynamic path optimization unit dynamically adjusts the permissions of the intercept coordination nodes through an iterative algorithm. The algorithm aims to minimize the deviation between dynamic intercept commands and the target drone's actual trajectory. It monitors the intermediate results output by each node in real time (such as the next-moment coordinates predicted by node A and the type of electromagnetic signal anomaly identified by node B) and compares them with the actual monitoring data. If a node's processing result deviates beyond a preset threshold (e.g., a spatial position error greater than 5 meters), the system automatically adjusts the node's permissions. For example, it could increase node A's processing priority for 3D coordinate data from the default 40% to 60%, while simultaneously reducing the corresponding permissions of other nodes. This adjustment process utilizes a heuristic search strategy, such as simulated annealing or particle swarm optimization, to iteratively try different permission combinations until it finds the configuration that minimizes the overall prediction error. For example, if the target drone enters a complex electromagnetic environment, causing fluctuations in electromagnetic signature data, the algorithm temporarily reduces node B's permissions and instead relies on the processing results of nodes A and C (responsible for attitude data) to generate intercept commands, minimizing interference.

[0047] The trajectory anomaly identification unit locates airspace threat zones based on compensated trajectory field data and electromagnetic coupling signature data. The system predefines threat zone types, including no-fly zones (e.g., over airports and government agencies), high-risk zones (e.g., near chemical plants and high-voltage power grids), and sensitive areas (e.g., crowded places). Each zone is assigned a corresponding spatial range and signature threshold. For example, a no-fly zone is defined as a spherical airspace with a radius of 500 meters centered at coordinates (x0, y0, z0). A spatial threat alert is triggered when the target drone's three-dimensional coordinate data enters this zone. High-risk zones are identified based on electromagnetic signature thresholds. For example, if industrial equipment communication signals of a specific frequency (e.g., 433MHz) are detected and the signal strength exceeds -60dBm, the target is considered to be near a chemical plant. The identification unit compares the target's trajectory with the threat zone boundary in real time and, based on whether the electromagnetic signature exceeds the threshold, generates a threat level assessment (e.g., low, medium, or high). For high-threat situations (e.g., a target entering a no-fly zone and electromagnetic signals indicating an out-of-control situation), the system immediately triggers an emergency interception process, generating dynamic intercept instructions with the shortest interception path and maximum speed instructions, driving the capture actuator to intercept at the fastest speed. For medium- and low-threat situations, a progressive interception strategy is adopted, such as first suppressing target communications through electromagnetic interference and then gradually adjusting the interception path. Throughout this process, the trajectory anomaly recognition unit continuously updates threat area information and works in conjunction with the dynamic path optimization unit to ensure that interception instructions are always generated for the most pressing threat scenario.

[0048] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0049] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based drone automatic capture system, characterized by: include: A dynamic environment perception module is used to acquire multimodal spatial data of the target UAV in real time. The multimodal spatial data includes a first monitoring sequence corresponding to three-dimensional coordinate data, a second monitoring sequence corresponding to flight attitude data, and a third monitoring sequence corresponding to electromagnetic signature data. The three-dimensional coordinate data includes first spatial positioning data generated by a multispectral radar and second dynamic trajectory data collected by a laser array sensor. A capture decision module is used to perform dynamic interference suppression processing on the multimodal spatial data and input the data into a decision processing layer for feature modeling, and generate dynamic interception instructions according to the output results of the decision processing layer; The decision processing layer includes a data cleaning module and a collaborative modeling module, wherein the data cleaning module is used to segment the trajectory of the original spatial data stream and filter the noise data. The collaborative modeling module is obtained by joint training based on the historical trajectory data and historical electromagnetic data of multiple interception scenarios; the collaborative modeling module includes an electromagnetic interference analysis layer, a trajectory correction layer and a decision execution layer connected in sequence.

2. The artificial intelligence-based drone automatic capture system according to claim 1, characterized in that: The electromagnetic interference analysis layer is used to perform time domain correlation processing on different monitoring sequences in the original spatial data stream to generate electromagnetic coupling feature data; the trajectory correction layer is used to model the dynamic trajectory correlation relationship between the electromagnetic coupling feature data corresponding to each monitoring sequence to generate compensation trajectory field data; The decision execution layer is used to generate dynamic interception instructions based on multi-dimensional fusion of compensation trajectory field data and electromagnetic coupling feature data.

3. The artificial intelligence-based drone automatic capture system according to claim 2, characterized in that: The modeling of the dynamic trajectory correlation relationship between the electromagnetic coupling characteristic data corresponding to each monitoring sequence to generate compensation trajectory field data includes: Using a trajectory optimization algorithm to identify key flight nodes in the electromagnetic coupling characteristic data, and determining a trajectory compensation sequence corresponding to each monitoring sequence based on the airspace type corresponding to each key flight node; Calculate the offset coefficient between the flight nodes in the same airspace in the trajectory compensation sequence corresponding to any two monitoring sequences, and generate the compensation trajectory field data between the any two monitoring sequences based on the offset coefficient.

4. The artificial intelligence-based drone automatic capture system according to claim 3, characterized in that: The calculation of the offset coefficient between flight nodes in the same airspace in the trajectory compensation sequence corresponding to any two monitoring sequences includes: When the number of flight nodes in the trajectory compensation sequence corresponding to any two monitoring sequences is inconsistent, virtual trajectory point interpolation is performed based on the airspace parameters corresponding to the terminal flight node in the one with fewer flight nodes, and the offset coefficient between the flight nodes in the same airspace is calculated based on the interpolated data.

5. The artificial intelligence-based drone automatic capture system according to claim 1, characterized in that: The data cleaning module is specifically used for: Dividing the first monitoring sequence, the second monitoring sequence, and the third monitoring sequence into equal density according to a preset spatial interval to generate standardized first spatial data, standardized second posture data, and standardized third electromagnetic data; A dynamic path clustering method is used to perform real-time calibration on the standardized first spatial data and the standardized second posture data, and a fixed spectrum analysis method is used to perform steady-state optimization on the standardized third electromagnetic data, and a first calibration sequence, a second calibration sequence, and a third calibration sequence are output; wherein the first calibration sequence includes the calibrated first spatial positioning data and the calibrated second dynamic trajectory data.

6. The artificial intelligence-based drone automatic capture system according to claim 5, characterized in that: The data cleaning module is also used for: Calculating a trajectory offset coefficient between the calibrated first spatial positioning data and the calibrated second dynamic trajectory data within a historical interception period; Predicting an expected trajectory value of the calibrated second dynamic trajectory data in the real-time interception period according to the trajectory deviation coefficient and the spatial parameters of the calibrated first spatial positioning data in the real-time interception period; Target interception compensation data is generated based on the calibrated second dynamic trajectory data and its expected trajectory value, and a monitoring sequence corresponding to the target interception compensation data is used as a first calibration sequence.

7. The artificial intelligence-based drone automatic capture system according to claim 2, characterized in that: The trajectory correction layer specifically includes: a path tracing unit, configured to perform trajectory path tracing for each monitoring sequence in the electromagnetic coupling characteristic data, so as to extract a corresponding trajectory propagation chain from each monitoring sequence; The spatial domain matching unit is used to perform spatiotemporal mapping between the trajectory propagation chain extracted from each monitoring sequence and the corresponding electromagnetic coupling feature data to generate compensated trajectory field data.

8. The artificial intelligence-based drone automatic capture system according to claim 7, characterized in that: The trajectory correction layer also includes: The signal anti-interference unit is used to perform electromagnetic hysteresis effect elimination processing on the compensation trajectory field data.

9. The artificial intelligence-based drone automatic capture system according to claim 2, characterized in that: The decision execution layer specifically includes: A multi-dimensional collaboration unit, comprising a plurality of interception collaboration nodes, each of which is connected to each monitoring sequence in the compensation trajectory field data and electromagnetic coupling characteristic data through permission configuration; a dynamic path optimization unit, configured to iteratively adjust the permission configuration using a dynamic path optimization algorithm to minimize the error between the dynamic interception instruction and the actual trajectory distribution; The trajectory anomaly recognition unit is used to locate the airspace threat area based on the compensated trajectory field data and the electromagnetic coupling characteristic data, and generate a dynamic interception instruction.

10. The artificial intelligence-based drone automatic capture system according to claim 5, characterized in that: The dynamic path clustering method specifically includes: Generate adaptive calibration parameters based on obstacle distribution characteristics of the real-time airspace environment; A spectrum analysis window mechanism is used to perform segmented calibration processing on the standardized first spatial data.

Images