Unmanned aerial vehicle jamming method based on multi-element information fusion mode of key defense site

Abstract

The application provides a kind of unmanned aerial vehicle interference method based on key defense site multi-element information fusion, belong to multi-element data fusion field, solve the problem such as multi-source heterogeneous data space-time alignment difficulty, low slow small target mode switching tracking divergence etc..The method comprises: collecting multi-source heterogeneous data and completing space-time alignment and coordinate conversion, obtaining target state observation value;Based on multi-source data, type identification is carried out, and target type and confidence are output;Adaptive Kalman filtering is carried out by interactive multi-model to predict state, and the optimal state estimation is obtained;In filter updating, noise covariance is adaptively adjusted according to observation residual, and consistency cross verification is carried out for different sensors to resist deception;According to target type, interference strategy library is matched, specific interference parameters are generated, target position after mechanical delay of rotating platform is predicted, azimuth and pitch angle are calculated, rotating platform is accurately pointed and interference signal is emitted.

Description

Technical Field

[0001] This invention belongs to the field of multi-data fusion, specifically a drone jamming method based on multi-data fusion of key defense locations. Background Technology

[0002] In recent years, civilian drone technology has developed rapidly, with diverse types and widespread applications in aerial photography, agricultural plant protection, power line inspection, and other fields, providing great convenience to the public. However, due to the difficulty of regulation and the continuously decreasing technical barriers, problems such as privacy violations and public safety incidents caused by drone misuse are becoming increasingly prominent, posing a serious challenge to the safety management of key areas.

[0003] Existing counter-drone systems mostly employ single-sensor devices, such as radar, wireless spectrum detection, or optoelectronic equipment. However, single sensors all have significant limitations in complex environments: radar has limited detection capabilities against low-altitude, slow-moving, and small targets and is susceptible to multipath interference; spectrum detection can only provide azimuth information and is easily deceived; and optoelectronic equipment is greatly affected by obstruction and lighting conditions. To address these issues, some technical solutions attempt to simply fuse data from multiple sensors, such as through direct weighted averaging or switching between them. However, these methods fail to effectively resolve the temporal asynchrony and spatial coordinate inconsistencies of multi-source heterogeneous data, resulting in insufficient accuracy of the fused target location.

[0004] Meanwhile, traditional Kalman filtering uses a single motion model, which cannot adapt to the rapid switching between hovering, constant speed, rapid acceleration, and sharp turns of UAVs. The state estimation error increases sharply during target maneuvers, even causing filter divergence and tracking loss. Furthermore, existing systems lack real-time detection and fault-tolerance mechanisms for sensor data anomalies. When a sensor is affected by environmental interference or malfunctions, erroneous data directly contaminates the fusion results, causing severe deviations in turntable pointing. Regarding the vulnerability of wireless spectrum equipment to deception interference, existing technologies lack multi-sensor cross-verification methods, making it difficult to effectively identify false target signals.

[0005] In the interference execution phase, existing methods typically employ a uniform interference strategy without differentiation based on UAV type. Furthermore, they operate in an open-loop control mode, lacking the ability to evaluate and adjust interference effects, resulting in low clearance efficiency when interference parameters do not match the target. Simultaneously, due to the inherent delay in turntable mechanical execution, directly using the current filtering position for pointing control can cause misses. The lack of a collaborative handover mechanism when deploying multiple turntables further reduces the success rate of continuous tracking and interference.

[0006] Therefore, there is an urgent need for a UAV jamming method that can achieve deep spatiotemporal alignment of multi-source heterogeneous data, adaptive tracking of maneuvering targets, sensor fault tolerance and anti-spoofing capabilities, and closed-loop optimization of jamming strategies. Summary of the Invention

[0007] This application provides a UAV jamming method based on multi-source information fusion in key defense locations, which solves the technical problems in the prior art, such as difficulty in spatiotemporal alignment of multi-source heterogeneous data, tracking divergence of low, slow and small targets when switching motion modes, filtering failure due to interference or failure of a single sensor, susceptibility of wireless spectrum equipment to deception interference affecting fusion reliability, low jamming efficiency due to mismatch between jamming strategy and UAV type, pointing lag and misses caused by turntable mechanical delay, and lack of effect evaluation and adaptive closed-loop adjustment for open-loop jamming.

[0008] To achieve the above objectives, this application adopts the following technical solution: Firstly, it provides a drone jamming method based on multi-source information fusion of key defense locations, including: Collect multi-source heterogeneous data of the target UAV, and perform spatiotemporal alignment and coordinate unification transformation on the multi-source heterogeneous data to obtain target state observation values; Based on the multi-source heterogeneous data, the target UAV is identified, and the target type and its confidence level are output. Based on the interactive multi-model adaptive Kalman filter algorithm, the target state observations are filtered and predicted to obtain the optimal state estimate of the target at the current moment. The optimal state estimate includes at least three-dimensional position and three-dimensional velocity. During the filtering update process, the observation noise covariance is adaptively adjusted based on the observation residuals of the acquisition devices or abnormal acquisition devices are shielded for a preset time. Consistency cross-validation is performed on the observation data between different acquisition devices to resist deceptive interference. Based on the target type and its confidence level, a pre-set interference strategy library is matched to generate targeted interference parameters. At the same time, based on the optimal state estimation, the target position after the mechanical delay of the turntable is predicted, the required horizontal azimuth and pitch angles of the turntable are calculated, and the turntable is controlled to rotate and transmit interference signals.

[0009] Based on the above technical solutions, the UAV jamming method based on multi-source information fusion in this application utilizes an interactive multi-model adaptive Kalman filter algorithm to fuse and predict multi-source heterogeneous data. This effectively addresses the rapid switching of UAVs between different motion modes such as hovering, constant speed, maneuvering, and turning, significantly improving tracking accuracy and stability under complex trajectories. The introduction of adaptive noise adjustment and abnormal sensor shielding mechanisms based on observation residuals maintains normal system operation even when a single sensor is interfered with or malfunctions, enhancing overall robustness. Furthermore, through cross-validation of consistency among multiple sensors, it effectively identifies… Furthermore, it resists deception and interference, enhancing the system's security in adversarial environments. By combining target type identification with a pre-set interference strategy library, it can adaptively match the optimal interference parameters based on the communication characteristics and anti-interference capabilities of different UAVs, avoiding the inefficiency caused by "one-size-fits-all" interference. It considers and predicts and compensates for the mechanical delay of the turntable, making the interference more precise and reducing misses caused by the lag of the actuator. Finally, the above-mentioned technical means work together to form a complete closed loop from multi-source perception, adaptive filtering, fault-tolerant fusion to strategy matching and precise interference, comprehensively improving the ability of key defense sites to detect, track, and counter "low, slow, and small" UAVs.

[0010] Furthermore, the collection of multi-source heterogeneous data from the target UAV includes: The polar coordinate data of the target UAV is acquired by millimeter-wave radar; the polar coordinate data includes at least range, azimuth angle and pitch angle. The radio signal characteristics of the UAV are collected by a wireless spectrum detection and direction-finding device; the radio signal characteristics include at least the center frequency, bandwidth, frequency hopping pattern, modulation method, and signal arrival direction; The visible light or infrared image of the target UAV and the corresponding WGS84 coordinate data are acquired by the photoelectric detection, tracking and ranging equipment to obtain image data containing coordinate information.

[0011] Furthermore, the spatiotemporal alignment and coordinate unification transformation of the multi-source heterogeneous data includes: Convert the polar coordinate data into latitude, longitude, and altitude data in WGS84 coordinates; Based on the timestamp of the millimeter-wave radar, the image data is interpolated and compensated using an adaptive sliding window cubic spline interpolation method, and the radio signal features are predicted and compensated using a Kalman smoother, resulting in compensated image data and compensated radio signal feature data. The latitude, longitude, and altitude data and the compensated data are synchronized using the NTP clock protocol to obtain the target state observation value.

[0012] Furthermore, the step of identifying the type of the target UAV based on the multi-source heterogeneous data includes: The radio signal features are compared with a pre-stored UAV spectrum database. The similarity of the spectrum features between each measured radio signal feature and the preset type of signal features is calculated using Mahalanobis distance or Euclidean distance. The similarity of the spectrum features is then normalized into a probability distribution using the Softmax function to obtain the first type of probability vector. The radar cross section and rotor micro-Doppler spectrum of the target UAV are extracted using millimeter-wave radar. The extracted features are then matched with a pre-stored UAV template. The cosine similarity between the feature vector and the pre-stored template is calculated and normalized to output the second type probability vector. The convolutional neural network extracts the outline, size, and rotor number features of the target drone from the image data. The convolutional neural network is pre-trained using an image dataset labeled with drone types. The output layer of the convolutional neural network is a Softmax classifier, which outputs probability values ​​for each category to obtain a third type probability vector. The first type probability vector, the second type probability vector, and the third type probability vector are fused using DS evidence theory to obtain a basic probability assignment for each candidate type. The type with the largest assignment is selected as the target type and the basic probability assignment is used as the confidence level. The target type includes consumer-grade, industrial-grade, racing drone, or unknown.

[0013] Furthermore, the interactive multi-model adaptive Kalman filter algorithm specifically includes: Step B1: Four motion modes are predefined, including hovering or low-speed drift mode, uniform linear motion mode, maneuver mode, and coordinated turning mode. Each mode corresponds to its own state transition matrix and basic process noise covariance matrix. Step B2, define the UAV's state vector as three-dimensional position and three-dimensional velocity: Where (x,y,z) represents the three-dimensional position of the UAV. Indicates the three-dimensional velocity of the drone; Step B3: Set the initialization parameters of the algorithm, including initial state estimation, initial covariance, initial probability of each motion mode, and mode transition probability matrix; Step B4, for each time point, perform the following sub-steps: Step B4-1: For each motion mode, calculate the mixing probability of each mode with the current mode based on the probability of each mode at the previous time step and the mode transition probability, and then calculate the mixing initial state estimate and mixing covariance of the current mode based on the mixing probability. Step B4-2: For each motion mode, the state transition matrix corresponding to the motion mode is used to predict the mixed initial state estimate to obtain the predicted state estimate, and the mixed covariance is predicted using the state transition matrix and the basic process noise covariance matrix to obtain the predicted covariance. Step B4-3: Obtain the target state observation value at the current moment, determine the observation matrix according to the observation source, and calculate the innovation vector and innovation covariance. Then calculate the Kalman gain, and use the Kalman gain to update the predicted state estimate and predicted covariance to obtain the updated state estimate and covariance under the current motion mode. Step B4-4: For each motion mode, calculate the likelihood function of the motion mode based on the innovation vector and innovation covariance, and then update the probability of the motion mode at the current time using the likelihood function, the probability of each mode at the previous time step, and the mode transition probability matrix. In steps B4-5, the updated state estimates of each mode are weighted and averaged according to the updated mode probabilities to obtain the final optimal state estimate. The updated covariances of each mode are then weighted and fused with the same weight to obtain the final covariance.

[0014] Furthermore, the filtering prediction process also includes dynamically adjusting process noise, specifically including: For each motion mode, obtain the updated mode probability of the motion mode in step B4-4, and determine the motion mode with the highest mode probability as the main mode at the current moment; For the main mode, the standardized innovation square is calculated based on the innovation vector and innovation covariance obtained in step B4-3; When the standardized innovation square is less than or equal to a preset threshold, it is determined that the main mode matches the actual motion state of the target UAV, and the basic process noise covariance matrix of the main mode remains unchanged. Otherwise, if the maneuverability of the target UAV exceeds the range of the main mode, the process noise covariance matrix of the main mode is increased according to the formula: ;in, This represents the adjusted process noise covariance matrix. This represents the basic process noise covariance matrix of the main mode. For the standardized innovation squared, This is a sensitivity parameter, with a value between 2 and 5, used to control the magnitude of noise increase; The adjusted process noise covariance matrix is ​​then used in step B4-2 for prediction at the next time step.

[0015] Furthermore, the adaptive adjustment of the observation noise covariance based on the observation residuals of the acquisition equipment includes: For each acquisition device s, before the filtering update in step B4-3, the standardized innovation square of the observation residual of the acquisition device s is calculated based on the innovation vector and innovation covariance matrix of the acquisition device; Based on the degrees of freedom of the standardized innovation squared distribution, a first anomaly threshold and a second anomaly threshold are set for the chi-square distribution of the observation dimension of the acquisition device; wherein, the second anomaly threshold is three times the first anomaly threshold, and the first anomaly threshold is used to judge mild anomalies, and the second anomaly threshold is used to judge severe anomalies. When the standardized innovation square is less than or equal to the first anomaly threshold, the data from the acquisition device is determined to be normal, and the observation noise covariance matrix of the acquisition device is maintained. When the square of the standardized innovation is greater than or equal to the second anomaly threshold, it is determined that the acquisition device has a severe anomaly, and the current update of the acquisition device is blocked. Only the predicted value is used as the state estimate at the current moment. At the same time, an anomaly counter is started for the acquisition device and the count is incremented by 1. Otherwise, the data acquisition device is deemed to have a minor malfunction, according to the formula. The observation noise covariance of the adaptive dilatation acquisition device; where... This represents the adjusted observation noise covariance. This represents the initial calibration value of the observation noise covariance matrix. For the standardized innovation squared, This is the first abnormal threshold; The adjusted observation noise covariance is substituted into the observation noise covariance matrix in the Kalman update to reduce the Kalman gain weight of the acquisition device in the update. When the fault counter of a certain data acquisition device reaches 3 consecutive times, the data acquisition device will be removed and will not be reconnected until it is manually re-inspected.

[0016] Furthermore, the cross-validation of consistency between observation data from different acquisition devices includes: Within each fusion cycle, based on the three-dimensional position in the optimal state estimate output by the interactive multi-model adaptive Kalman filter algorithm, and combined with the known installation position of the wireless spectrum detection and direction finding device, the theoretical azimuth angle from the spectrum device to the target UAV is calculated. Obtain the measured azimuth angles between the wireless spectrum detection and direction finding device and the target UAV; According to the formula Calculate the theoretical azimuth angle and the measured azimuth angle angular deviation ; The state of the millimeter-wave radar equipment and the photoelectric detection, tracking and ranging equipment is determined based on the standardized innovation square. When the standardized innovation square is less than or equal to the first anomaly threshold, it is determined that a reliable reference exists and the consistency between the radar equipment and the photoelectric equipment is marked as good. If the angle deviation exceeds a preset angle threshold and the consistency between the radar equipment and the optoelectronic equipment is good, it is determined that the wireless spectrum detection and direction finding equipment has received deception interference or has malfunctioned, and the fusion weight factor of the wireless spectrum detection and direction finding equipment is set to a preset minimum value. The fusion weight factor is used in the Kalman filter update process. When using the observation data of the wireless spectrum detection and direction finding equipment, the observation noise covariance matrix of the wireless spectrum detection and direction finding equipment is divided by the fusion weight factor to obtain the adjusted observation noise covariance matrix, which is then substituted into the Kalman gain calculation formula to reduce the impact of equipment observations on state estimation updates. If the angle deviation recovers to within the preset angle threshold for M consecutive cycles, the value of the fusion weight factor is restored to the calibration value with a step size of 0.1 per cycle; where M is a positive integer greater than 1.

[0017] Furthermore, the method also includes continuously tracking the state changes of the target UAV during and after the UAV jamming operation, calculating the quantitative evaluation index of the jamming effect, and adjusting the jamming strategy and filtering parameters based on the closed-loop feedback of the evaluation results.

[0018] Furthermore, the calculation of quantitative evaluation indicators for interference effects, and the adjustment of interference strategies and filtering parameters based on closed-loop feedback of evaluation results, include: Obtain the distance, speed, and altitude of the current target drone relative to the defense center to obtain the current target motion status data; Calculate the distance change rate, speed decay rate, and altitude change rate of the target UAV within a preset period based on the current target motion state data; The Sigmoid function is used to map the negative values ​​of the distance change rate, the velocity decay rate, and the altitude change rate to the [0,1] interval, and the three mapped values ​​are weighted and summed to obtain the interference effect score. If the interference effect score is lower than the first threshold and the duration exceeds the first time, the current interference power will be increased by 3 dB. If the current interference power has reached the maximum power, the alternative interference mode will be switched. If the interference effect score is lower than the second threshold and the duration exceeds the second time, an interference failure alarm will be triggered and a backup interference source will be activated. If the interference effect score is higher than the third threshold and the distance relative to the defense center is greater than the safe distance, then keep the current interference parameters unchanged; If the interference effect score is higher than the fourth threshold and the distance relative to the defense center is less than the success distance, the interference is considered successful and the interference operation is terminated. The interference effect evaluation results are fed back to the interactive multi-model adaptive Kalman filter algorithm. When the rate of change of the interference effect score exceeds the preset value and the target motion pattern changes abruptly, the transition coefficient from the current mode to the maneuvering mode in the mode transition probability matrix is ​​temporarily increased, and the original value is restored after a preset time.

[0019] Furthermore, the step of generating targeted interference parameters by matching a pre-set interference strategy library based on the target type and its confidence level includes: Construct an interference strategy library, which includes at least the following fields: target type, confidence threshold, preferred interference method, preferred interference parameters, alternative interference methods, alternative interference parameters, and switching conditions; wherein, the interference parameters include at least the frequency, power, modulation method, transmitted waveform, and duration of the interference signal; When the confidence level is greater than or equal to the preset confidence threshold, the corresponding preferred interference method and preferred interference parameters are directly matched according to the target type; otherwise, wideband frequency sweep interference is adopted as the default strategy, and the frequency sweep range covers 800MHz to 6GHz, in 10MHz increments. During the interference process, if the effectiveness score of the current interference method is lower than the preset score threshold for a period of time exceeding the preset duration, the method will be switched from the preferred interference method to the alternative interference method. The effectiveness score of the interference method is calculated based on the rate of change of the distance, speed and altitude of the target UAV relative to the defense center.

[0020] Furthermore, the interference strategies for each target type in the interference strategy library are as follows: Consumer-grade: The preferred jamming method is GPS L1 band navigation decoy, with an output power of 10mW and a decoy signal format that simulates real GPS ephemeris, causing the target to mistakenly believe that its position has deviated and automatically return to home; the alternative jamming method is 2.4GHz / 5.8GHz communication link blocking, with a power of 1W. Industrial grade: The preferred jamming method is multi-band communication blocking, with a power of 5W; the alternative jamming method is navigation decoy + high-power microwave pulse, with a microwave peak power of 10kW. Racing drone: The preferred jamming method is remote control link blocking, using a directional high-gain antenna with a power of 2W; the alternative jamming method is video link blocking with a power of 1W. Unknown type: Wideband sweep frequency interference is used, with adjustable power. The initial power is set to 1W and dynamically adjusted according to the subsequent interference effect score S.

[0021] Further, the step of predicting the target position after mechanical delay of the turntable based on the optimal state estimation, calculating the required horizontal azimuth and pitch angles of the turntable, controlling the turntable rotation, and transmitting interference signals includes: Obtain the target position and target velocity from the optimal state estimate at the current moment; The target position is predicted and compensated based on the mechanical delay time of the turntable: The predicted and compensated target location is obtained; where, This represents the target velocity vector at the current time k. This represents the target position in the optimal state estimate at time k. The target acceleration estimate at the current time k is obtained by the state difference of the current master mode filter, and τ represents the mechanical delay time of the turntable; Transform the predicted and compensated target position into the station center coordinate system with the fixed point of the turntable as the origin, and calculate the relative coordinates and horizontal distance. Calculate the horizontal azimuth and pitch angles of the target UAV; wherein, the horizontal azimuth is the angle rotated clockwise from due north to the target direction, and is calculated using the arctangent function based on the relative coordinates; the pitch angle is the angle positive upwards with the horizontal plane as zero degrees, and is calculated using the arctangent function based on the ratio of the vertical coordinate to the horizontal distance; The calculated horizontal azimuth and pitch angles are sent to the turntable drive unit via the communication interface to control the turntable to rotate to the specified angle. Once the turntable has rotated to its position, an interference signal is transmitted according to the matched interference parameters. When deploying multiple turntables, the turntable closest to the target and with the smallest absolute pitch angle is selected to perform interference. At the same time, the time when the target will leave is predicted based on the coverage of the current turntable. When the remaining coverage time is less than one second, the optimal state estimate and predicted trajectory of the current target are sent to the adjacent turntables, so that the adjacent turntables can rotate to the predicted handover point in advance, thus achieving seamless handover of the interference task.

[0022] Compared with the prior art, the beneficial effects of this application are: Firstly, in terms of multi-source data acquisition and processing, this application integrates three heterogeneous sensors: millimeter-wave radar, wireless spectrum detection, and photoelectric detection. Through unified coordinate transformation, high-precision time synchronization, and adaptive interpolation and prediction compensation for different sampling frequencies, it effectively solves the spatiotemporal inconsistency problem of multi-source data, providing high-quality, well-aligned target observations for subsequent fusion. Based on this, utilizing three layers of information—spectral features, radar cross-section and micro-Doppler spectrum, and image features—combined with template matching, neural networks, and DS evidence theory, decision-level fusion is performed. This enables accurate identification of different types of UAVs, such as consumer-grade, industrial-grade, and racing drones, and outputs confidence scores, thus providing a reliable basis for differentiated jamming strategies and significantly improving the classification capability for low, slow, and small targets in complex environments.

[0023] Secondly, regarding state estimation and fault-tolerant fusion, this application designs an interactive multi-model adaptive Kalman filter algorithm. It predefines four typical motion modes and updates the mode probabilities in real time, enabling adaptive following of the UAV's rapid switching between hovering, constant speed, maneuvering, and turning states, significantly improving the accuracy and stability of trajectory tracking. Simultaneously, by calculating the standardized innovation square of the observation residuals and setting dual thresholds, it achieves adaptive observation noise expansion for minor anomalies in sensor data and automatic shielding for severe anomalies, effectively preventing filter divergence caused by single sensor failures or brief interference. Furthermore, utilizing consistent cross-validation between radar and photoelectric systems, it can detect and suppress deceptive interference that may affect wireless spectrum equipment. By dynamically adjusting the fusion weighting factor, it reduces the impact of unreliable observations, significantly enhancing the system's robustness and security in adversarial environments.

[0024] Finally, regarding the jamming strategy and execution loop, this application matches a pre-set jamming strategy library based on the target type and its confidence level, providing differentiated jamming methods such as navigation deception, communication disruption, and microwave pulses for different UAVs. When the confidence level is low, a wideband frequency sweep conservative strategy is automatically adopted, achieving adaptive optimization of jamming parameters. During the jamming process, changes in target distance, speed, and altitude are continuously tracked, and the jamming effect score is quantitatively calculated. Power is dynamically adjusted or the jamming mode is switched based on the score, and the evaluation results are fed back to the filter. When the target escapes, the mode transition probability is temporarily increased, forming a perception effect. Fusion interference Evaluate The complete closed-loop system was readjusted. Simultaneously, position prediction compensation was performed to address the mechanical delay of the turntable, and seamless handover was achieved when multiple turntables were deployed, ensuring precise and continuous jamming targeting. These technical measures worked synergistically to comprehensively enhance the detection, tracking, identification, and countermeasure capabilities against UAVs in key defense locations, effectively addressing the shortcomings of existing systems in maneuver tracking, anti-jamming, type adaptation, and closed-loop control. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating the UAV jamming method based on multi-source information fusion of key defense locations provided in this application embodiment; Figure 2 A flowchart illustrating another UAV jamming method based on multi-source information fusion of key defense locations, provided in an embodiment of this application; Figure 3 A flowchart illustrating another UAV jamming method based on multi-source information fusion of key defense locations, provided in an embodiment of this application; Figure 4 This is a flowchart illustrating another drone jamming method based on multi-source information fusion of key defense locations, provided as an embodiment of this application. Detailed Implementation

[0027] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0028] To address the technical problems in existing technologies, such as the susceptibility of single sensors to environmental interference leading to inaccurate target detection, the inability to identify UAV types resulting in a lack of targeted jamming strategies, and the inability of traditional filtering methods to adapt to changes in UAV maneuvers leading to low tracking accuracy, embodiments of this application provide a UAV jamming method based on multi-source information fusion of key defense locations, such as... Figure 1 As shown, the method includes: Step S1: Collect multi-source heterogeneous data of the target UAV, and perform spatiotemporal alignment and coordinate unification transformation on the multi-source heterogeneous data to obtain target state observation values ​​that are time-synchronized and have consistent coordinate systems.

[0029] The multi-source heterogeneous data can include radar detection data, radio spectrum monitoring data, image data acquired by electro-optical tracking equipment, and acoustic fingerprint data obtained by acoustic sensor arrays. Because these data come from different sources, have different sampling frequencies, and different coordinate references, time alignment must be achieved through interpolation or resampling, and coordinate transformation must be used to convert the data from each sensor to the same geographic coordinate system or the local coordinate system of the defense site before they can be fused to form consistent state observations. This step aims to eliminate the inherent differences between sensors, providing accurate and comparable input for subsequent filtering and fusion, which is the foundation of multi-source information fusion. Common methods for achieving spatiotemporal alignment and unified coordinate transformation include using the Global Positioning System (GPS) or Network Time Protocol (NAT) to provide a unified timestamp for each sensor, and using cubic spline interpolation to align non-uniformly spaced data to the same time axis. Coordinate transformation involves converting radar polar coordinates, electro-optical device image pixel coordinates, and acoustic array spherical coordinates to the Northeast-Northeast coordinate system or the universal transverse Mercator projection coordinate system, with transformation parameters determined by the installation and calibration position of each sensor at the defense site.

[0030] Step S2: Based on the spectral characteristics, radar cross section characteristics, and photoelectric image characteristics in the multi-source heterogeneous data, the target UAV is identified, and the target type and its confidence level are output.

[0031] Among these features, spectral characteristics refer to the frequency, bandwidth, and modulation method of the drone's image or data transmission signals; radar cross-section characteristics reflect the drone's size, shape, and materials; and photoelectric image characteristics include the drone's external outline, number of rotors, and color. Fusion of these three features significantly improves identification accuracy and prevents deception by decoys or similar targets based on a single feature. This step determines the specific drone model, such as the DJI Phantom series, Mavic series, or a custom-made racing drone, thus providing a basis for matching the interference strategy in step S6. Common drone type identification methods include pre-constructing a feature database of typical drone types and using classifiers such as support vector machines, random forests, or convolutional neural networks for multimodal feature fusion identification. The final output is the identification result and confidence score, which can be obtained through the probability value output by the Softmax function or the classifier's voting ratio.

[0032] Step S3: Based on the interactive multi-model adaptive Kalman filter algorithm, filter and predict the target state observations, identify the current motion mode of the target UAV online and dynamically adjust the process noise to obtain the optimal state estimate of the target UAV at the current moment.

[0033] The optimal state estimation includes at least three-dimensional position and three-dimensional velocity. The Interactive Multiple Model Adaptive Kalman Filter (IMM-AKF) is an improvement on the classic IMM-KF. The classic IMM-KF uses a fixed set of motion models, such as uniform velocity, uniform acceleration, and cooperative turning models, with the process noise covariance of each model preset unchanged. When the actual UAV maneuvering does not match the model, the tracking error increases. The adaptive improvement is reflected in the introduction of a real-time model probability update mechanism, simultaneously estimating the process noise covariance of each model online. This can be achieved, for example, through a covariance matching method based on residuals within a sliding window or an adaptive factor based on maximum likelihood estimation, enabling the filter to quickly respond to sudden acceleration, sharp turns, and other maneuvering changes by the UAV. This step overcomes the limitation of a single motion model in describing maneuvering targets, improving the tracking accuracy and stability of highly maneuverable UAVs. In terms of specific implementation, three to five motion models can be set up in the interactive multi-model framework. Each model uses capacitive Kalman filtering or unscented Kalman filtering to process nonlinear observations. Process noise adaptation can be achieved by Sage-Husa adaptive filtering or by adjusting the noise amplitude using fuzzy logic.

[0034] In step S4, during the filtering update process, the observation residuals of each sensor are statistically checked in real time. When the sensor data is determined to be abnormal, the observation noise covariance of the sensor is adaptively adjusted or the abnormal sensor is temporarily shielded. Consistency cross-validation is performed on the observation data between different sensors to resist deceptive interference.

[0035] In this step, the sensors refer to the various devices used in step S1 to collect multi-source heterogeneous data, such as radar, spectrum analyzers, and photoelectric trackers. The basic principle of real-time statistical testing to determine data anomalies is that, when the filter is working normally, the observation residual of each sensor—the difference between the actual observed value and the predicted value—should follow a Gaussian distribution with zero mean, and its standardized innovation squared should follow a chi-square distribution. When a sensor is interfered with, blocked, or malfunctions, the mean of the residuals will deviate from zero or the variance will increase significantly. By calculating the chi-square test statistic of the innovation sequence and comparing it with a preset threshold, if it exceeds the threshold, the data is determined to be abnormal. The adaptive adjustment of the sensor observation noise covariance reduces the weight of abnormal data in state updates, thereby avoiding filter divergence; while temporarily shielding abnormal sensors completely isolates their data during periods of continuous anomaly until subsequent tests restore normalcy. In addition, one way to implement consistency cross-validation to resist deception interference is to compare the target distance measured by radar with the distance calculated by photoelectric image through triangulation, or to analyze the difference between the UAV signal arrival angle obtained by spectrum monitoring and the azimuth angle measured by radar. If the difference exceeds the preset tolerance threshold, it is determined that there may be deception interference. At this time, the fusion center will give priority to the observations consistent by the majority of sensors or enable the redundant sensor voting mechanism.

[0036] Step S5: During and after the jamming operation, continuously track the state changes of the target UAV, calculate the quantitative evaluation index of the jamming effect, and adjust the jamming strategy and filtering parameters based on the closed-loop feedback of the evaluation results.

[0037] The core function of this step is to form closed-loop control, avoiding energy waste or defense vulnerabilities caused by blind interference. The calculation of the quantitative evaluation index of interference effect can be based on the following parameters: the rate of change of distance between the target UAV's three-dimensional position and the core area of ​​the defense site (i.e., approach speed); the change in the yaw angle of the target's three-dimensional velocity vector to reflect whether the interference causes it to deviate from its original course; and the change in target type confidence, such as the disappearance of spectral characteristics or a decrease in identification confidence after successful interference. In addition, indicators for whether the target exhibits uncontrolled hovering, returning to base, or crashing can be introduced. One way to adjust the interference strategy and filtering parameters based on the evaluation results is as follows: if the evaluation score is below a threshold (i.e., the effect is poor), switch the interference waveform, such as from continuous wave to pulse wave, increase the interference power, or switch from suppressive interference to deceptive interference; if the target exhibits violent maneuvers, increase the process noise covariance of the adaptive Kalman filter in step S3 to enhance tracking agility; if the target tends to stabilize, appropriately reduce the process noise to improve filtering smoothness.

[0038] Step S6: Match the preset interference strategy library according to the target type and its confidence level to generate targeted interference parameters. At the same time, predict the target position after the mechanical delay of the turntable based on the optimal state estimation, calculate the required horizontal azimuth and pitch angles of the turntable, control the turntable to rotate and transmit interference signals.

[0039] The method for generating targeted interference parameters involves utilizing the optimal interference frequency, waveform, modulation method, power level, and illumination time corresponding to different drone models pre-stored in the interference strategy library. For example, frequency-hopping tracking interference is used in the 2.4 GHz and 5.8 GHz bands for DJI Phantom series, while wideband suppression interference is used for racing drones. The target position mentioned in this step refers to the predicted drone position after the turntable's mechanical delay time. Because there is a mechanical delay between the turntable receiving the command and actually rotating into position, the optimal state estimate at the current moment cannot be used for direct aiming. Instead, the position after a delay time needs to be predicted by extrapolating the three-dimensional position and three-dimensional velocity included in the current optimal state estimate. The overall purpose of this step is to ensure that the interference signal can accurately point to the future position of the drone, compensate for the turntable's inertial delay, improve the probability of interference hit, and dynamically optimize the interference parameters according to the drone type to achieve highly efficient "targeted" interference.

[0040] Based on this, the UAV jamming method based on multi-source information fusion of key defense locations provided in this embodiment can integrate heterogeneous data from multiple sources such as radar, spectrum, and optoelectronics to overcome the shortcomings of single sensors being susceptible to environmental or deceptive interference; it achieves stable tracking of highly maneuverable UAVs through interactive multi-model adaptive Kalman filtering; it uses type recognition to match targeted jamming strategies to avoid blind jamming; and it continuously optimizes through closed-loop feedback of jamming effects, which can improve the success rate of intercepting intruding UAVs and the security of defense locations.

[0041] In one possible implementation of this application embodiment, the above-mentioned S1 can be specifically implemented by the following S101, S102 and S103, which are described in detail below: S101. Collect multi-source heterogeneous data of the target UAV through millimeter-wave radar, wireless spectrum detection and direction finding equipment and photoelectric detection, tracking and ranging equipment respectively.

[0042] Among them, the millimeter-wave radar is used to transmit high-frequency electromagnetic waves and receive target echoes. By measuring the time difference and phase difference of the echoes, the distance, azimuth, and elevation angles of the target relative to the radar are obtained, forming polar coordinate data. The wireless spectrum detection and direction-finding equipment is used to passively monitor the communication link between the UAV and the remote controller, extracting radio signal characteristics, including center frequency, bandwidth, frequency hopping pattern, modulation method, and signal arrival direction. The photoelectric detection, tracking, and ranging equipment is used to acquire visible light or infrared images of the target UAV, and uses the built-in laser ranging module or binocular stereo vision algorithm to obtain the WGS84 coordinate data corresponding to the target, thereby obtaining image data containing coordinate information.

[0043] For example, after deploying the above three devices at a certain defense site, when a DJI Phantom 4 drone enters the warning area, the radar will continuously output a set of polar coordinate sequences, the spectrum device will capture the OFDM modulation signal in the 2.4GHz band, and the optoelectronic device will collect image sequences containing the quadcopter outline and the corresponding latitude, longitude and altitude data.

[0044] S102. Spatiotemporal alignment of the collected multi-source heterogeneous data, including using the millimeter-wave radar timestamp as a reference, employing an adaptive sliding window cubic spline interpolation method to interpolate and compensate the data of the photoelectric detection tracking and ranging equipment, and using a Kalman smoother to predict and compensate the data of the wireless spectrum detection and direction finding equipment, to obtain compensated image data and compensated radio signal feature data.

[0045] The purpose of spatiotemporal alignment is to eliminate time synchronization issues caused by different sampling frequencies among devices. Since millimeter-wave radar has the highest sampling frequency and is most sensitive to dynamic targets, its timestamp is used as the reference.

[0046] In some implementations, for optoelectronic devices with relatively low sampling frequencies, an adaptive sliding window cubic spline interpolation method is used. At each radar sampling moment, a continuous and smooth cubic spline curve is fitted based on several preceding and following optoelectronic observations. The target position corresponding to the radar moment is then interpolated onto this curve. The size of the sliding window is dynamically adjusted according to the target's maneuverability: when the target's motion mode is determined to be uniform or hovering, the window can be expanded to 0.5 seconds to smooth noise; when the target is maneuvering, the window shrinks to 0.1 seconds to preserve details.

[0047] For spectrum devices with extremely low sampling frequencies, simple interpolation is ineffective due to the sparse data and nonlinear variations. Therefore, a Kalman smoother is used for prediction compensation. The Kalman smoother first establishes a state-space model using historical spectrum data, then performs forward filtering and backward smoothing for each radar moment, outputting an estimated value of the spectrum characteristics at that moment.

[0048] It should be noted that after spatiotemporal alignment, the data from all devices are converted to a unified radar timeline. Each radar sampling moment corresponds to a complete set of radar, photoelectric, and spectrum data. However, the spectrum data is obtained through prediction compensation, and its reliability is slightly lower than that of the measured data. In subsequent fusion, it can be weighted by the observation noise covariance.

[0049] S103. Transform the spatiotemporally aligned data into the WGS84 coordinate system and synchronize the time using the NTP clock protocol to obtain the target state observation values.

[0050] The coordinate unification transformation refers to converting the polar coordinate data of the millimeter-wave radar, as well as the local coordinate data of the optoelectronic and spectrum equipment, to the WGS84 Earth ellipsoidal coordinate system, so that all data are expressed under the same spatial reference. The specific method for millimeter-wave radar polar coordinate transformation is as follows: First, the polar coordinate data (range ρ, azimuth θ, elevation φ) is converted to a geocentric rectangular coordinate system with the radar center as the origin. The conversion formula is: Where X, Y, and Z represent the rectangular coordinates of the target in the Earth-centered Earth-fixed coordinate system.

[0051] Then, the rectangular coordinates are converted to WGS84 latitude, longitude, and altitude coordinates (longitude L, latitude B, altitude H) using an iterative algorithm. The iterative formula is as follows: Where a is the Earth's major semi-axis, b is the Earth's minor semi-axis, e is the first eccentricity, and e' is the second eccentricity. Convergence is typically achieved after 3 to 4 iterations.

[0052] For optoelectronic devices, the WGS84 coordinates they directly output do not require conversion; for spectrum devices, their direction finding results are usually azimuth angles with the device itself as the origin. Combined with the device's own WGS84 coordinates, they can be converted into the target's geodetic coordinates through geodetic theme calculation methods.

[0053] After completing the coordinate transformation, it is also necessary to synchronize the time of all devices through the NTP clock protocol to ensure that the timestamp of each data point is traced back to the same clock source, with a synchronization accuracy better than 1 millisecond.

[0054] In some implementations, the NTP clock protocol can deploy an NTP time server within the defense site. This server acquires high-precision UTC time via GPS or BeiDou receivers. All sensor devices perform time calibration with the NTP server via a network, with a calibration cycle of once per second. After coordinate transformation and time synchronization, the data from each radar moment forms a complete set of "target status observations," which include three-dimensional position (longitude, latitude, and altitude) and optional velocity information (provided by Doppler velocity from optoelectronic devices or radar).

[0055] It should be noted that when a spectrum device cannot be converted into complete three-dimensional coordinates due to a lack of distance information, its azimuth angle can be retained as an independent observation dimension, and partial observation updates can be performed through the observation matrix H in subsequent Kalman filter updates.

[0056] Based on the above technical solution, this embodiment realizes accurate spatiotemporal alignment and unified coordinate transformation of multi-source heterogeneous data, providing reliable and consistent target state observations for subsequent type identification, adaptive filtering and interference strategies, thus laying the data foundation for the entire UAV interference method.

[0057] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, the above S2 can be implemented through the following S201, S202 and S203, which are explained in detail below: S201. The radio signal features collected by the wireless spectrum detection and direction finding device are compared with the pre-stored UAV spectrum database. The similarity of the spectrum features between each measured signal feature and the preset type signal features is calculated using Mahalanobis distance or Euclidean distance. The similarity is then normalized into a probability distribution using the Softmax function to obtain the first type probability vector.

[0058] The radio signal characteristics include at least the center frequency, bandwidth, frequency hopping pattern, modulation method, and signal direction of arrival. The pre-stored drone spectrum database is established by collecting, analyzing, and labeling the remote control and image transmission signals of various typical drones (such as DJI Phantom, DJI Mavic, Autel, and racing drones) under standard test environments. Each record contains the above-mentioned characteristic parameters and their corresponding drone type label.

[0059] When using Mahalanobis distance to calculate similarity, considering that the features have different dimensions and are correlated, the Mahalanobis distance is weighted by the inverse of the covariance matrix on the eigenvectors. The calculation formula is as follows: ;in The measured feature vectors, Let be the feature mean vector of the i-th type of UAV in the database. This is the covariance matrix of this type of feature. The smaller the Mahalanobis distance, the higher the similarity.

[0060] If Euclidean distance is used, the formula is: .

[0061] After obtaining the distances for each type, they need to be converted into similarity scores, usually by taking the reciprocal or negative exponent of the distance, and then normalizing them into a probability distribution using the Softmax function. ;in This represents the probability that the target belongs to the i-th class, where N is the total number of drone types pre-stored in the database. This yields the first class probability vector. .

[0062] In some implementations, to accommodate database updates following the emergence of new drones, the pre-stored spectrum database should adopt an scalable structure, supporting the incremental addition of new types. Simultaneously, since the spectrum characteristics of the same drone model may differ under different flight modes, the database can store multiple feature templates for typical operating states for each type, selecting the optimal matching result during comparison.

[0063] It should be noted that the results of spectrum feature comparison depend on signal quality. When the target UAV is in silent flight and does not emit any radio signals, the spectrum device cannot collect effective features. In this case, the first type probability vector should be marked as "unavailable". In the subsequent fusion, only radar and photoelectric features are used for type identification.

[0064] S202. Use millimeter-wave radar to extract the radar cross section value and rotor micro-Doppler spectrum of the target UAV. Perform template matching between the extracted features and the pre-stored UAV template, calculate the cosine similarity between the feature vector and the pre-stored template, and output the second type probability vector after normalization.

[0065] The radar cross section (RCS) reflects a target's ability to reflect radar waves and is related to the target's size, shape, material, and attitude; its unit is decibel-square meter. The rotor micro-Doppler spectrum is the spectral structure formed by the periodic frequency modulation of the radar echo caused by the high-speed rotation of the rotor. Different UAVs have different rotor numbers, rotational speeds, and blade shapes, resulting in variations in the characteristic frequencies and spacing of their micro-Doppler spectra. Millimeter-wave radar extracts the micro-Doppler spectrum through long-term coherent accumulation and time-frequency analysis (such as short-time Fourier transform).

[0066] The pre-stored UAV template library is established by measuring and eigenvectorizing the RCS and micro-Doppler spectra of known UAVs in the same frequency band. Each template contains a feature vector. During template matching, the measured feature vectors are used... With the i-th template vector Perform cosine similarity calculation. Obtain the cosine similarity of each template. Then, it is normalized and transformed into a probability distribution: Where M is the total number of templates, and the maximum value function is used to set the negative similarity to zero. This yields the second-type probability vector. .

[0067] In some implementations, to improve matching robustness, a multi-feature fusion template matching strategy can be adopted. This involves simultaneously using the RCS mean, RCS fluctuation statistics (standard deviation), micro-Doppler spectrum dominant and harmonic positions, spectral line spacing, etc., to construct a high-dimensional feature vector. Furthermore, since target attitude changes affect RCS values, the template library should contain RCS templates of the same UAV at different azimuth angles. During matching, the corresponding template subset is selected based on the target's current azimuth angle.

[0068] It should be noted that the rotor micro-Doppler spectrum extracted by radar is easily affected by the target speed and radar operating mode. When the target is moving at high speed, the micro-Doppler spectrum may be broadened or blurred. In this case, the fusion weight of radar features can be appropriately reduced.

[0069] S203. Extract the outline, size, and rotor number features of the target UAV from the image data collected by the photoelectric detection tracking and ranging device using a convolutional neural network. The convolutional neural network is pre-trained using an image dataset labeled with UAV types. The network output layer is a Softmax classifier, which outputs the probability values ​​of each category to obtain the third type probability vector.

[0070] Convolutional Neural Networks (CNNs) are deep learning models that automatically learn hierarchical features from images. The CNN structure used in this step can be a lightweight network such as MobileNetV2 or ShuffleNet to meet the real-time requirements of edge computing devices. The network's input layer receives visible light or infrared images (after scaling and normalization preprocessing) acquired by optoelectronic devices. The intermediate layers progressively extract low-level features (edges, corners) and high-level semantic features (outlines, rotors, fuselage shapes) through convolution, pooling, and activation operations. Finally, the output layer, consisting of fully connected layers and a Softmax layer, yields the probability of each predefined type. During network training, a large-scale image dataset labeled with drone types is used, with cross-entropy loss as the loss function and Adam as the optimizer. Multiple iterations are performed until convergence. After training, real-time acquired images are input into the network for a single forward propagation to obtain the third-type probability vector. , where K is the number of types.

[0071] In some implementations, to improve recognition accuracy, a multi-frame image feature fusion strategy can be adopted. This involves temporally smoothing the CNN output probabilities of consecutive frames (e.g., by moving average) to suppress misjudgments caused by noise in a single frame or brief occlusion of the target. Furthermore, when the optoelectronic device operates in infrared mode, the CNN should use weight parameters specifically trained for thermal infrared images, because the contours and thermal distribution of targets in infrared images differ significantly from those in visible light images.

[0072] S204. Using DS evidence theory, the first type probability vector, the second type probability vector, and the third type probability vector are fused to obtain the basic probability assignment for each candidate type. The type with the largest assignment is selected as the target type and the basic probability assignment is used as the confidence level.

[0073] The target types include consumer-grade, industrial-grade, racing drones, or unknown types. DS evidence theory is a fusion method for handling uncertain information, effectively addressing conflicts between different pieces of evidence. Specific operational steps may include: First, each probability vector is considered as a basic probability assignment function (BPA) for independent evidence sources. Let the recognition framework... ={Consumer-grade, Industrial-grade, Racing Drone, Unknown}, where the basic probability of each evidence source (spectrum, radar, photoelectric) for each subset A is assigned a value. And satisfy .

[0074] For the first, second, and third probability vectors, they can be directly used as the corresponding evidence sources to assign values ​​to single-point types, while simultaneously assigning a smaller probability to the entire set. This indicates uncertainty, that is: ;in The uncertainty of the evidence source can be dynamically set according to the current operating state of the sensor (e.g., when the spectrum device has no signal). Set to 0.9).

[0075] Then, the Dempster combination rule is used to fuse two sources of evidence; for three sources of evidence, they are fused pairwise sequentially. The fusion formula is: ;in, K is the conflict coefficient, representing the degree of conflict between two sources of evidence. When the conflict is too large (e.g., K is close to 1), conflict correction strategies can be introduced, such as redistributing the conflict probability to the entire set or using a weighted average fusion.

[0076] After fusion, the basic probability assignment for each candidate type is obtained. And the uncertainty of the entire series The type with the highest assigned basic probability is selected as the final target type for output, and this basic probability is used as the confidence level. If the basic probability assigned to all types is less than 0.5 and the uncertainty is high, then "Unknown" is output.

[0077] In some implementations, to improve the real-time performance of fusion, a fusion result table for each evidence source combination can be pre-calculated, allowing for direct table lookup at runtime. Furthermore, when the fusion result is stable over multiple consecutive time points, time smoothing or state machine constraints can be employed to prevent frequent jumps in type recognition results.

[0078] It should be noted that the advantage of DS evidence fusion is that it can explicitly express uncertainty. When a sensor fails completely, its uncertainty can be set to 1, and the fusion result will depend entirely on the other sensors.

[0079] Based on the above technical solution, this embodiment extracts probability vectors from three heterogeneous features: spectrum, radar, and photoelectric, and uses DS evidence theory for decision-level fusion. This can accurately identify the specific type of UAV and its confidence level, providing a reliable basis for matching subsequent differentiated interference strategies. At the same time, the robustness and anti-deception capability of the identification system are enhanced through the evidence conflict handling mechanism.

[0080] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 3 As shown, the above S3 can be implemented through the following S301, S302 and S303, which are explained in detail below: S301. Predefine four motion modes and set initialization parameters, including state vector, state transition matrix, basic process noise covariance matrix, initial probability of each mode, and mode transition probability matrix.

[0081] The four motion modes are: Mode 1, hovering or low-speed drifting mode, suitable for drones that are almost stationary or drifting slowly in the air; Mode 2, uniform linear motion mode, suitable for drones flying in a straight line at a constant speed; Mode 3, maneuvering mode, suitable for drones performing violent maneuvers such as rapid acceleration, rapid deceleration, or uncoordinated turns; and Mode 4, coordinated turning mode, suitable for drones turning at a constant angular velocity. Each mode corresponds to its own state transition matrix and basic process noise covariance matrix.

[0082] The state vector is defined as three-dimensional position and three-dimensional velocity: Where (x,y,z) represents the three-dimensional position of the UAV in the WGS84 coordinate system. This represents the corresponding three-dimensional velocity.

[0083] State transition matrix This is used to describe the change in state from time k-1 to time k under mode i. For a uniform motion model, Given a block diagonal matrix, the transition submatrix on each coordinate axis is: For the hovering model, the velocity transfer coefficient can be set to zero; for the coordinated turning model, the turning angular velocity needs to be introduced as an additional state. Basic process noise covariance matrix. This reflects the uncertainty in model predictions, the magnitude of which is related to model characteristics, particularly the maneuvering mode. Typically larger than hover mode.

[0084] Initialization parameters also include: initial state estimation (This can be obtained from the observations of the first few frames through least squares fitting), initial covariance (Reflecting the uncertainty of the initial estimate), initial probabilities of each motion mode and mode transition probability matrix ,in Let represent the probability of transitioning from pattern i to pattern j, and satisfy . .

[0085] In some implementations, the mode transition probability matrix can be adjusted based on historical experience or machine learning methods to adapt to the target characteristics of different defense locations. For example, in airport airspace, drones mostly move at a constant speed, so the probability of transitioning from a constant speed mode to a maneuvering mode is set relatively low; while over important facilities, drones may hover for reconnaissance, so the probability of self-transition from a hovering mode is set relatively high.

[0086] S302. For each time step, perform the recursive steps of the interactive multi-model Kalman filter, including input interaction, prediction, update, mode probability update, and output fusion.

[0087] This step is the core of interactive multi-model Kalman filtering, and it specifically includes the following sub-steps: Sub-step S302-1 (Input Interaction): For each motion mode j, based on the probabilities of each mode in the previous time step... and mode transition probability Calculate the mixture probability of each mode in the previous time step with respect to the current mode j: ; Then calculate the mixed initial state estimate of mode j. and mixed initial covariance : ; ; in, This represents the state estimate of mode i at the previous time step. Let represent the covariance matrix of mode i at the previous time step; Sub-step S302-2 (prediction): For each mode j, use the state transition matrix of that mode. Make a prediction: ; ; in, This indicates that pattern j is the current state estimate obtained from the mixed state prediction of the previous time step. This represents the prediction covariance matrix of pattern j. The underlying process noise covariance matrix of pattern j; Sub-step S302-3 (Update): Obtain the target state observation value at the current time. (Obtained from step S1), determine the observation matrix based on the observation source. For example, when the observation is a three-dimensional position, When the observation is only the azimuth angle, Since it is a nonlinear function, linearization using extended Kalman filtering is required. Calculate the innovation vector. New covariance matrix and Kalman gain matrix : ; ; ; Then update the state estimate. Covariance : ; ;in, It is the identity matrix; Sub-step S302-4 (Mode Probability Update): For each mode j, calculate the likelihood function. , where m is the dimension of the observation vector, Let represent the determinant of the new information covariance matrix. Then update the mode probabilities: ;in, Let j be the probability of the updated pattern j; Sub-step S302-5 (output fusion): The updated state estimates of each mode are weighted and averaged according to the updated mode probabilities to obtain the final optimal state estimate. : ; The final covariance matrix Also merged with the same weight: .

[0088] In some implementations, to reduce computational complexity, a pattern can be temporarily removed when its probability is less than a certain threshold (such as 0.01) and then reactivated at a later time.

[0089] It should be noted that the observation matrix The form depends on the type of sensor: when using polar coordinate observations from millimeter-wave radar, the observation equation is nonlinear and should be linearized using extended Kalman filtering or unscented Kalman filtering; when using three-dimensional position observations after coordinate transformation, the observation equation is linear.

[0090] S303. Dynamically adjust the process noise during the filtering prediction process, including determining the main mode at the current moment, calculating the standardized innovation square, and adaptively increasing the process noise covariance matrix of the main mode according to the relationship between the innovation square and the preset threshold.

[0091] First, for each motion pattern, obtain its updated pattern probability in S302-4. The pattern with the highest probability is determined as the dominant pattern at the current moment. .

[0092] For this main mode, the innovation vector calculated in S302-3 is used. and new information covariance Calculate the standardized innovation square: This statistic follows a chi-square distribution with degrees of freedom equal to the dimension m of the observation vector.

[0093] Set preset threshold The 99th percentile of the chi-square distribution (e.g., when m=3). ).when At that time, it is determined that the kinematic assumptions of the main mode match the actual motion state of the target, while maintaining the basic process noise covariance matrix of the main mode. constant.

[0094] when When the target UAV's maneuver intensity exceeds the assumptions of the main mode, meaning the actual maneuver is more intense than the model anticipates, it's necessary to temporarily increase the process noise to enhance the filter's tracking capability. The adjustment formula is: ;in This represents the adjusted process noise covariance matrix. This represents the basic process noise covariance matrix of the main mode. To standardize the square of the new information, This is a sensitivity parameter, with a value between 2 and 5. The physical meaning is to scale the degree of exceeding the threshold according to the sensitivity, and then take the maximum value of the sum of the two values ​​to ensure that the noise is not reduced.

[0095] The adjusted process noise covariance matrix is ​​substituted into S302-2 for prediction of the next time step (i.e., replacing the original). To smooth out changes, you can set the time limit to be set for multiple consecutive time periods. After restoring to within the threshold, gradually increase... decay back For example, multiply by 0.9 every moment.

[0096] In some implementations, the squared innovations of multiple modes can be monitored simultaneously, but usually adjusting only the main mode is sufficient. Additionally, to prevent excessive process noise from causing filter instability, [the following can be done]: Set an upper limit, for example, a maximum of 10. It should be noted that the normalized innovation square is not only used for process noise adjustment, but also for sensor anomaly detection in the subsequent step S4.

[0097] Based on the above technical solution, this embodiment uses an interactive multi-model adaptive Kalman filter algorithm to identify the current motion mode of the UAV online and dynamically adjust the process noise. This effectively solves the problem of tracking divergence of a single model when the target is maneuvering, significantly improves the filtering accuracy and stability under complex trajectories, and provides accurate target state estimation for subsequent turntable pointing and interference strategies.

[0098] In one possible implementation of this application embodiment, the above-mentioned S4 specifically includes the following S401 to S403: S401. Before updating the Kalman filter, calculate the standardized innovation square of the observation residual for each acquisition device, set the thresholds for mild anomalies and severe anomalies according to the chi-square distribution, and determine the device status based on the interval in which the standardized innovation square is located. For mild anomalies, adaptively expand the observation noise covariance; for severe anomalies, shield the device from this update.

[0099] The data acquisition device s includes a millimeter-wave radar, a wireless spectrum detection and direction-finding device, and an electro-optical detection, tracking, and ranging device. Before the filtering update in step S3, the information vector of this device is acquired. Calculate the standardized innovation square: This statistic follows a set of degrees of freedom. The chi-square distribution, where Let s be the observation dimension of device s (3 for radar and photoelectric, 2 for spectrum).

[0100] Set the first exception threshold The 99th percentile of the chi-square distribution, the second outlier threshold The first threshold is used to assess mild abnormalities, and the second threshold is used to assess severe abnormalities. According to... The value of is handled in three cases: when If the equipment data is deemed normal, the observed noise covariance matrix of the equipment is kept at its calibration value. constant.

[0101] when If a device is determined to have a severe anomaly, its update for that device is directly disabled: that is, during the update phase in step S3, the device's observation data is not used, only the predicted value is used as the state estimate, and an anomaly counter is started for that device. And add 1.

[0102] when When a minor malfunction is detected in the equipment, it is determined according to the formula. Adaptive dilation observation noise covariance. In this formula, This indicates the multiple by which the anomaly exceeds the normal threshold. This multiple is multiplied by the calibration noise, increasing the observation noise and thus automatically reducing the weight of the device's observations in subsequent Kalman gain calculations. The adjusted... Substitute the original formula into the update formula of step S3. .

[0103] In some implementations, to prevent frequent state jumps caused by brief data jitter, a check can be performed before determining whether a minor or major anomaly is detected. Perform smoothing processing on multiple consecutive frames, for example, taking the three most recent frames. The average value is used as the basis for judgment. In addition, for spectrum devices, due to their fewer observation dimensions, smaller degrees of freedom of the chi-square distribution, and relatively lower threshold values, they are more likely to trigger anomaly judgments. Therefore, the threshold can be appropriately relaxed (e.g., using the 95th percentile) to reduce false judgments.

[0104] It should be noted that the calculation of the normalized innovation square depends on the innovation covariance of the Kalman filter. When the sensor data quality deteriorates, the innovation covariance will increase accordingly. It is a normalized metric that allows for fair comparisons between different sensors.

[0105] S402. For acquisition devices that are repeatedly identified as severely abnormal, temporarily remove them from the fusion architecture until they pass manual re-inspection or system self-inspection before reconnecting them.

[0106] Each acquisition device s maintains an anomaly counter. The initial value is 0. Whenever a severe anomaly is determined in S401 (i.e., ...), the device is considered to have experienced a serious anomaly. When the device is identified as severely abnormal for multiple consecutive moments, the counter increments by 1. Once the device returns to normal operation... The counter is immediately reset to zero.

[0107] When a device's anomaly counter counts three times consecutively, the device is determined to have experienced a persistent malfunction or severe interference, and it is temporarily removed from the converged architecture. After removal, subsequent Kalman filter updates will no longer use the device's data, relying solely on other functioning devices for tracking. Simultaneously, the system generates an alarm message, prompting maintenance personnel to conduct a manual re-inspection. Once the manual re-inspection confirms the device has returned to normal, or the system confirms the device's availability through a self-test procedure (such as sending test signals to verify device response), the device's anomaly counter is reset to zero, and it is reconnected to the converged architecture.

[0108] In some implementations, to maintain a certain level of redundancy during device removal, backup sensors or a purely predictive mode can be employed. For critical defense sites, multiple sensors of the same type (such as two radars) can be deployed, allowing one to take over when the other is removed. It should be noted that removal operations only apply to persistent severe anomalies; minor anomalies will not trigger removal, as they may be caused by target maneuvers or transient environmental interference, which can be effectively suppressed through adaptive expansion noise without the need for complete shielding.

[0109] S403. In each fusion cycle, the consistency cross-validation of radar and optoelectronic equipment is used to detect whether the wireless spectrum detection direction finding equipment is subject to deception interference. If the angle deviation exceeds the threshold and the radar optoelectronic consistency is good, the fusion weight factor of the spectrum equipment is reduced, and its influence is weakened by adjusting the observation noise covariance in the subsequent Kalman update.

[0110] In each fusion cycle (i.e., each radar sampling time), the three-dimensional position is first determined based on the optimal state estimate output in step S3. Combined with the known installation locations of wireless spectrum detection and direction finding equipment Calculate the theoretical azimuth angle from the spectrum device to the target: ;in The function returns the angle of rotation clockwise from true north, with a range of values ​​of 1000. .

[0111] Simultaneously acquire the azimuth angle measured by the spectrum analyzer. Calculate the angle deviation: This formula ensures that the deviation is always the minimum included angle.

[0112] Then, the state of the millimeter-wave radar and optoelectronic equipment is determined based on the normalized innovation square calculated in S401: if both... All are less than or equal to their respective first anomaly thresholds If a reliable reference exists, the consistency between the marked radar equipment and optoelectronic equipment is good; if If the angle exceeds a preset threshold (e.g., 10°) and the radar and photoelectric sensors are in good agreement, it is determined that the wireless spectrum detection and direction finding equipment has been subjected to deception interference or has malfunctioned, and its weighting factors are then fused. Set to a preset minimum value (e.g., 0.1). This fusion weighting factor is used in subsequent Kalman filter updates: when using observation data from a spectrum analyzer, divide its observation noise covariance matrix by... ,Right now This is equivalent to amplifying observation noise (because) <1, thereby reducing the impact of the device's observations on the state estimation update.

[0113] If the angle deviation recovers to within the preset angle threshold for M consecutive cycles (M can be 10), then the angle deviation will be adjusted in a step size of 0.1 per cycle. Gradually restore to the calibration value of 1.0.

[0114] It should be noted that consistency cross-validation is only used to detect whether spectrum devices are being spoofed, and not to correct anomalies in radar or optoelectronic devices, because radar and optoelectronic devices have already performed self-diagnosis through residual detection in S401.

[0115] Based on the above technical solutions, this embodiment achieves sensor-level data quality monitoring and fault tolerance through residual detection and adaptive noise adjustment. It effectively identifies and suppresses deceptive interference through consistent cross-validation, improves the robustness and security of the multi-source fusion system in complex electromagnetic environments and adversarial scenarios, and provides reliable target state estimation for subsequent interference execution.

[0116] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 4 As shown, the above S5 specifically includes the following S501 to S503: S501. During and after the jamming operation, continuously track the state changes of the target UAV, obtain the current distance, speed and altitude of the target relative to the defense center, and calculate the distance change rate, speed decay rate and altitude change rate within a preset period.

[0117] The jamming execution period refers to the entire process from the initial transmission of the jamming signal to the determination of successful jamming or termination. Continuous tracking refers to using the optimal state estimate output from step S3 to acquire the target's current three-dimensional position in each fusion cycle (typically the sampling cycle of a millimeter-wave radar, such as 0.01 seconds). and three-dimensional velocity Define the location of the defense center or the fixed point of the turntable as follows: Then, the distance of target k relative to the defense center at the current moment is: ; The target speed is: The target height is... .

[0118] Set evaluation cycle =1 second, and the interference effect index is calculated once every 5 evaluation cycles (i.e., every 5 seconds) to ensure statistical stability.

[0119] Calculate the rate of change of distance: ;in The distance was 5 seconds ago. The denominator is the time interval between adjacent moments (e.g., 0.01 seconds). The time interval is 5 seconds. A positive rate of change in distance indicates that the target is moving away from the defense center, which is in line with the expected interference; a negative value indicates that the target is moving closer.

[0120] Similarly, velocity decay rate: A negative velocity decay rate indicates that the target is decelerating, which is in line with expectations; a positive value indicates acceleration.

[0121] Rate of change in height: A negative rate of change in altitude indicates a descent (forced landing), which is in line with expectations; a positive value indicates an ascent.

[0122] It should be noted that the distance change rate, velocity decay rate, and altitude change rate can be calculated based on the smoothed values ​​after filtering rather than the original estimates, in order to reduce the impact of noise. Furthermore, each of these three rates independently reflects a different dimension of the interference's effect: the distance change rate focuses on safety protection, the velocity decay rate focuses on suppression effects, and the altitude change rate focuses on forced landing induction.

[0123] S502. The Sigmoid function is used to map the negatives of the distance change rate, the velocity decay rate, and the altitude change rate to the [0,1] interval, and the three mapped values ​​are weighted and summed to obtain the interference effect score. The interference power is dynamically adjusted or the interference mode is switched based on the comparison result between the score and the preset threshold.

[0124] The Sigmoid function has the following form: Its output ranges from 0 to 1, and it can smoothly map any real value to a probabilistic score. This is due to the distance rate of change. The larger (more positive) the value, the farther away the target is, and the better the effect. Therefore, it can be used directly. Regarding speed, the desired speed decay rate is... Since it is a negative value, we take the opposite number. Mapping, i.e. ,when When it is negative, A positive value results in a higher score; similarly, the rate of change of height is taken as... .

[0125] The overall interference effect score is defined as follows: ;in For the preset weighting coefficients, satisfy The default value is The value of S ranges from [0,1], with values ​​closer to 1 indicating better interference. The following thresholds and strategies are set: If S < 0.3 and the duration exceeds 2 seconds, the current interference power will be increased by 3dB (equivalent to doubling the power); if the current power has reached the maximum allowable power of the device, the alternative interference mode will be switched.

[0126] If S < 0.1 and the duration exceeds 5 seconds, an interference failure alarm will be triggered, and a backup interference source (such as a high-power microwave device) will be activated.

[0127] If S > 0.8 and the target distance is... (If the safe distance is 500 meters), then the current interference parameters remain unchanged.

[0128] If S > 0.9 and the target distance is... If the successful interference is within a certain distance (e.g., 100 meters), the interference is considered successful and the interference operation is terminated.

[0129] It should be noted that the duration determination is to avoid erroneous operations caused by instantaneous fluctuations. Furthermore, since the interference effect score can quantify the multi-dimensional changes in target distance, speed, and altitude into a unified numerical indicator, the system can automatically determine the effectiveness of the current interference strategy and execute closed-loop control operations such as power adjustment, mode switching, or success determination. This avoids energy waste caused by blindly continuing interference and allows for rapid response when the target escapes, significantly improving the interference success rate and the utilization efficiency of defense resources.

[0130] S503. Feedback the interference effect evaluation results to the interactive multi-model adaptive Kalman filter algorithm. When the rate of change of the interference effect score exceeds the preset value and the target motion mode changes abruptly, temporarily increase the transition coefficient from the current mode to the maneuvering mode in the mode transition probability matrix, and restore the original value after a preset time.

[0131] The interference effect evaluation results are not only used to adjust interference parameters, but also fed back to the filtering algorithm to improve tracking performance. Specifically, after calculating the interference effect score S in each evaluation cycle (every 5 seconds), its rate of change is also calculated: ;in The score from the previous assessment, =5 seconds.

[0132] At the same time, the current mode probability distribution is obtained from the interactive multi-model output of step S3 to determine whether a mode mutation has occurred, that is, whether the current master mode has changed from the master mode of the previous time.

[0133] When detected When the score decreases at a rate exceeding 0.1 every 5 seconds (equivalent to a decrease of 0.1 every 5 seconds) and a sudden change in mode occurs (e.g., from a constant speed mode to a maneuvering mode), it is determined that the target has taken a violent escape maneuver after being interfered with. In this case, the mode transition probability matrix is ​​temporarily increased. The transition coefficient from the current mode to the maneuver mode (Mode 3). For example, the transition coefficient from the current mode to the maneuver mode (Mode 3). (Probability of self-transfer in maneuver mode) increases by 0.2, while decreasing accordingly. The value of (j≠3) is kept at a row sum of 1. This adjustment makes the filter more inclined to switch to maneuver mode in subsequent moments, thus responding more quickly to the target's escape maneuvers. This temporary adjustment lasts for a preset time (e.g., 5 seconds) before reverting to the original state. matrix.

[0134] Based on the above technical solution, this embodiment continuously tracks the changes in the target's state during the interference process, quantifies and calculates the interference effect score, and dynamically adjusts the interference power or switches the interference mode according to the score. At the same time, the evaluation results are fed back to the filtering algorithm to optimize the tracking performance, forming a complete closed-loop control, which improves the interference success rate and adaptability to target escape maneuvers.

[0135] In one possible implementation of this application embodiment, S6 specifically includes the following S601 to S603: S601. Based on the target type and its confidence level, match the preset interference strategy library to generate targeted interference parameters, including the preferred interference method and alternative interference methods and their parameters. When the confidence level is lower than the threshold, use wideband frequency sweep interference as the default strategy.

[0136] The pre-configured interference strategy library is a data table containing at least the following fields: target type, confidence threshold, preferred interference method, preferred interference parameters, alternative interference methods, alternative interference parameters, and switching conditions. Target types include consumer-grade, industrial-grade, racing drone, or unknown, and are output by step S2.

[0137] The confidence level c is the basic probability assigned to the S2 output, and its value ranges from [0,1].

[0138] The preset reliability threshold can be 0.7. When When c < 0.7, the preferred interference method and parameters are directly matched according to the target type; when c < 0.7, wideband sweep interference is adopted as the default strategy, with a sweep range covering 800MHz to 6GHz, in 10MHz increments, and an initial power of 1W. The interference parameters include at least the frequency, power, modulation method, transmitted waveform, and duration of the interference signal.

[0139] The specific strategies for each target type are as follows: Consumer-grade: The preferred jamming method is GPS L1 band (1575.42MHz) navigation decoy, with parameters of 10mW output power and decoy signal format simulating real GPS ephemeris, causing the target to mistakenly believe that its position has deviated and automatically return to home; the alternative jamming method is 2.4GHz / 5.8GHz communication link blocking, with a power of 1W.

[0140] Industrial grade: The preferred jamming method is multi-band communication blocking, covering 900MHz, 2.4GHz, 5.8GHz and 4G / 5G bands, with a power of 5W; the alternative jamming method is navigation decoy + high-power microwave pulse, with a microwave peak power of 10kW.

[0141] For racing drones: the preferred jamming method is remote control link blocking, using a directional high-gain antenna at a frequency of 2.4GHz or 900MHz (depending on the commonly used remote control frequency band), with a power of 2W; the alternative jamming method is video link blocking (5.8GHz), with a power of 1W.

[0142] Unknown type: Wideband sweep frequency interference is used, with adjustable power. It is initially set to 1W and dynamically adjusted according to the interference effect score in subsequent step S5.

[0143] During the interference process, if the effect score of the current interference method (calculated by step S5) is lower than the preset score threshold (e.g., 0.3) and the duration exceeds the preset duration (e.g., 10 seconds), then the preferred interference method is switched to the alternative interference method.

[0144] S602. Based on the optimal state estimate at the current moment, predict the target position after the mechanical delay of the turntable, calculate the required horizontal azimuth and pitch angles of the turntable, and control the turntable to rotate to the specified angle.

[0145] Specifically, the target position in the optimal state estimate at the current time k output in step S3 is obtained. and target speed Turntable mechanical delay time The delay is determined by the turntable model and load, typically ranging from 50ms to 200ms, and can be obtained through calibration experiments. To compensate for this delay, prediction... Target location after the time limit: ;in The target acceleration estimate at the current moment is obtained by the state difference of the current master mode filter, i.e. .

[0146] The predicted target position is converted to a fixed point on the turntable. Using a station-centered coordinate system (northeast-northeast coordinate system) as the origin, calculate the relative coordinates: ; Horizontal distance Horizontal azimuth The formula for calculating the angle of clockwise rotation from due north to the target direction is: ;in The range of function return values ​​is It needs to be converted to 0° to 360°: if the result is less than 0, add 360°.

[0147] Pitch angle Angles with zero degrees above the horizontal plane and positive upwards: ; The calculated and The signal is sent to the turntable drive unit via a communication interface (such as RS422 or Ethernet) to control the turntable to rotate to the specified angle. After the turntable has rotated to the designated position (which can be confirmed by reading feedback from the turntable encoder), an interference signal is emitted based on the interference parameters generated by S601.

[0148] In some implementations, closed-loop correction can be introduced to further improve pointing accuracy: the photoelectric device captures the deviation between the interference signal illumination spot and the target position in real time, calculates the miss distance through image processing, and feeds it back to the turntable for fine-tuning. It should be noted that the acceleration term... It is crucial in scenarios involving highly maneuverable targets; neglecting it could lead to significant prediction errors.

[0149] S603. When deploying multiple turntables, select the turntable closest to the target and with the smallest absolute pitch angle to perform interference, and send the target status to the adjacent turntables to achieve seamless handover when the target is about to leave the coverage area of ​​the current turntable.

[0150] Multiple turntables are deployed at different locations within the defense area. Each turntable has a limited horizontal sector coverage range (e.g., ±60°) and an elevation sector coverage range (e.g., -20° to +60°). At each instant, the system calculates the distance of the target relative to each turntable based on the predicted target position calculated by S602. and pitch angle The selection of a turntable for jamming must meet two conditions: first, the target must be within the turntable's coverage area (i.e., both the horizontal azimuth and elevation angles must be within the sector); second, the selection criteria must be... Minimum and Minimum (priority can be set to distance as primary and pitch as secondary).

[0151] After determining which turntable to execute the interference, the azimuth and elevation angles of that turntable, calculated in S602, are sent to that turntable to execute the interference. Simultaneously, the rate of change of the target's horizontal azimuth angle relative to the current turntable is continuously monitored to estimate the remaining coverage time. ;in To ensure the turntable horizontally covers the boundary angle (e.g., 60°). This is the current absolute azimuth. The azimuth rate of change is obtained by dividing the azimuth difference between two consecutive frames by time.

[0152] when When the time is less than 1 second, the system sends the optimal state estimate of the current target to the adjacent turntable (whose coverage area is adjacent to the current turntable). ,speed And the predicted trajectory (i.e., the position at several future moments). After receiving the information, adjacent turntables rotate in advance to the predicted azimuth and elevation angles of the handover point, achieving a seamless handover of the interference mission.

[0153] In some implementations, a master-slave cooperative strategy can be adopted: the master relay station is responsible for tracking and jamming; when the target is about to leave its coverage area, the master relay station sends a handover command to the slave relay station, and the master relay station stops jamming after the slave relay station locks onto the target. To avoid jamming interruption, a "connect first, disconnect later" approach can be used, that is, the slave relay station transmits jamming first, and then the master relay station stops.

[0154] Based on the above technical solutions, this embodiment generates targeted interference parameters through type matching, ensures accurate turntable pointing through mechanical delay prediction compensation, and achieves continuous coverage through multi-turntable collaboration, thereby significantly improving the success rate of interference and continuous tracking capability. It effectively solves the technical problems of low efficiency of single interference strategies, turntable lag and misses, and discontinuous handover of multiple turntables.

[0155] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0156] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely illustrative descriptions of the application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A drone jamming method based on multi-source information fusion of key defense locations, characterized in that, include: Collect multi-source heterogeneous data of the target UAV, and perform spatiotemporal alignment and coordinate unification transformation on the multi-source heterogeneous data to obtain target state observation values; Based on the multi-source heterogeneous data, the target UAV is identified, and the target type and its confidence level are output. Based on the interactive multi-model adaptive Kalman filter algorithm, the target state observations are filtered and predicted to obtain the optimal state estimate of the target at the current moment. The optimal state estimate includes at least three-dimensional position and three-dimensional velocity. During the filtering update process, the observation noise covariance is adaptively adjusted based on the observation residuals of the acquisition devices or abnormal acquisition devices are shielded for a preset time. Consistency cross-validation is performed on the observation data between different acquisition devices to resist deceptive interference. Based on the target type and its confidence level, a pre-set interference strategy library is matched to generate targeted interference parameters. At the same time, based on the optimal state estimation, the target position after the mechanical delay of the turntable is predicted, the required horizontal azimuth and pitch angles of the turntable are calculated, and the turntable is controlled to rotate and transmit interference signals.

2. The UAV jamming method based on multi-source information fusion of key defense locations as described in claim 1, characterized in that, The multi-source heterogeneous data collected from the target UAV includes: The polar coordinate data of the target UAV is acquired by millimeter-wave radar; the polar coordinate data includes at least range, azimuth angle and pitch angle. The radio signal characteristics of the UAV are collected by a wireless spectrum detection and direction-finding device; the radio signal characteristics include at least the center frequency, bandwidth, frequency hopping pattern, modulation method, and signal arrival direction; The visible light or infrared image of the target UAV and the corresponding WGS84 coordinate data are acquired by the photoelectric detection, tracking and ranging equipment to obtain image data containing coordinate information.

3. The UAV jamming method based on multi-source information fusion of key defense locations as described in claim 2, characterized in that, The spatiotemporal alignment and coordinate unification transformation of the multi-source heterogeneous data includes: Convert the polar coordinate data into latitude, longitude, and altitude data in WGS84 coordinates; Based on the timestamp of the millimeter-wave radar, the image data is interpolated and compensated using an adaptive sliding window cubic spline interpolation method, and the radio signal features are predicted and compensated using a Kalman smoother, resulting in compensated image data and compensated radio signal feature data. The latitude, longitude, and altitude data and the compensated data are synchronized using the NTP clock protocol to obtain the target state observation value.

4. The UAV jamming method based on multi-source information fusion of key defense locations as described in claim 2, characterized in that, The step of identifying the type of the target UAV based on the multi-source heterogeneous data includes: The radio signal features are compared with a pre-stored UAV spectrum database. The similarity of the spectrum features between each measured radio signal feature and the preset type of signal features is calculated using Mahalanobis distance or Euclidean distance. The similarity of the spectrum features is then normalized into a probability distribution using the Softmax function to obtain the first type of probability vector. The radar cross section and rotor micro-Doppler spectrum of the target UAV are extracted using millimeter-wave radar. The extracted features are then matched with a pre-stored UAV template. The cosine similarity between the feature vector and the pre-stored template is calculated and normalized to output the second type probability vector. The convolutional neural network extracts the outline, size, and rotor number features of the target drone from the image data. The convolutional neural network is pre-trained using an image dataset labeled with drone types. The output layer of the convolutional neural network is a Softmax classifier, which outputs probability values ​​for each category to obtain a third type probability vector. The first type probability vector, the second type probability vector, and the third type probability vector are fused using DS evidence theory to obtain a basic probability assignment for each candidate type. The type with the largest assignment is selected as the target type and the basic probability assignment is used as the confidence level. The target type includes consumer-grade, industrial-grade, racing drone, or unknown.

5. The UAV jamming method based on multi-source information fusion of key defense locations according to claim 1, characterized in that, The interactive multi-model adaptive Kalman filter algorithm specifically includes: Step B1: Four motion modes are predefined, including hovering or low-speed drift mode, uniform linear motion mode, maneuver mode, and coordinated turning mode. Each mode corresponds to its own state transition matrix and basic process noise covariance matrix. Step B2, define the UAV's state vector as three-dimensional position and three-dimensional velocity: Where (x,y,z) represents the three-dimensional position of the UAV. Indicates the three-dimensional velocity of the drone; Step B3: Set the initialization parameters of the algorithm, including initial state estimation, initial covariance, initial probability of each motion mode, and mode transition probability matrix; Step B4, for each time point, perform the following sub-steps: Step B4-1: For each motion mode, calculate the mixing probability of each mode with the current mode based on the probability of each mode at the previous time step and the mode transition probability, and then calculate the mixing initial state estimate and mixing covariance of the current mode based on the mixing probability. Step B4-2: For each motion mode, the state transition matrix corresponding to the motion mode is used to predict the mixed initial state estimate to obtain the predicted state estimate, and the mixed covariance is predicted using the state transition matrix and the basic process noise covariance matrix to obtain the predicted covariance. Step B4-3: Obtain the target state observation value at the current moment, determine the observation matrix according to the observation source, and calculate the innovation vector and innovation covariance. Then calculate the Kalman gain, and use the Kalman gain to update the predicted state estimate and predicted covariance to obtain the updated state estimate and covariance under the current motion mode. Step B4-4: For each motion mode, calculate the likelihood function of the motion mode based on the innovation vector and innovation covariance, and then update the probability of the motion mode at the current time using the likelihood function, the probability of each mode at the previous time step, and the mode transition probability matrix. In steps B4-5, the updated state estimates of each mode are weighted and averaged according to the updated mode probabilities to obtain the final optimal state estimate. The updated covariances of each mode are then weighted and fused with the same weight to obtain the final covariance.

6. The UAV jamming method based on multi-source information fusion of key defense locations according to claim 5, characterized in that, The filtering prediction process also includes dynamically adjusting process noise, specifically including: For each motion mode, obtain the updated mode probability of the motion mode in step B4-4, and determine the motion mode with the highest mode probability as the main mode at the current moment; For the main mode, the standardized innovation square is calculated based on the innovation vector and innovation covariance obtained in step B4-3; When the standardized innovation square is less than or equal to a preset threshold, it is determined that the main mode matches the actual motion state of the target UAV, and the basic process noise covariance matrix of the main mode remains unchanged. Otherwise, if the maneuverability of the target UAV exceeds the range of the main mode, the process noise covariance matrix of the main mode is increased according to the formula: ;in, This represents the adjusted process noise covariance matrix. This represents the basic process noise covariance matrix of the main mode. For the standardized innovation squared, This is a sensitivity parameter, with a value between 2 and 5, used to control the magnitude of noise increase; The adjusted process noise covariance matrix is ​​then used in step B4-2 for prediction at the next time step.

7. The UAV jamming method based on multi-source information fusion of key defense locations as described in claim 5, characterized in that, The adaptive adjustment of the observation noise covariance based on the observation residuals of the acquisition equipment includes: For each acquisition device s, before the filtering update in step B4-3, the standardized innovation square of the observation residual of the acquisition device s is calculated based on the innovation vector and innovation covariance matrix of the acquisition device; Based on the degrees of freedom of the standardized innovation squared distribution, a first anomaly threshold and a second anomaly threshold are set for the chi-square distribution of the observation dimension of the acquisition device; wherein, the second anomaly threshold is three times the first anomaly threshold, and the first anomaly threshold is used to judge mild anomalies, and the second anomaly threshold is used to judge severe anomalies. When the standardized innovation square is less than or equal to the first anomaly threshold, the data from the acquisition device is determined to be normal, and the observation noise covariance matrix of the acquisition device is maintained. When the square of the standardized innovation is greater than or equal to the second anomaly threshold, it is determined that the acquisition device has a severe anomaly, and the current update of the acquisition device is blocked. Only the predicted value is used as the state estimate at the current moment. At the same time, an anomaly counter is started for the acquisition device and the count is incremented by 1. Otherwise, the data acquisition device is deemed to have a minor malfunction, according to the formula. The observation noise covariance of the adaptive dilatation acquisition device; where... This represents the adjusted observation noise covariance. This represents the initial calibration value of the observation noise covariance matrix. For the standardized innovation squared, This is the first abnormal threshold; The adjusted observation noise covariance is substituted into the observation noise covariance matrix in the Kalman update to reduce the Kalman gain weight of the acquisition device in the update. When the fault counter of a certain data acquisition device reaches 3 consecutive times, the data acquisition device will be removed and will not be reconnected until it is manually re-inspected.

8. The UAV jamming method based on multi-source information fusion of key defense locations according to claim 7, characterized in that, The cross-validation of consistency of observation data between different acquisition devices includes: Within each fusion cycle, based on the three-dimensional position in the optimal state estimate output by the interactive multi-model adaptive Kalman filter algorithm, and combined with the known installation position of the wireless spectrum detection and direction finding device, the theoretical azimuth angle from the spectrum device to the target UAV is calculated. Obtain the measured azimuth angles between the wireless spectrum detection and direction finding device and the target UAV; According to the formula Calculate the theoretical azimuth angle and the measured azimuth angle angular deviation ; The state of the millimeter-wave radar equipment and the photoelectric detection, tracking and ranging equipment is determined based on the standardized innovation square. When the standardized innovation square is less than or equal to the first anomaly threshold, it is determined that a reliable reference exists and the consistency between the radar equipment and the photoelectric equipment is marked as good. If the angle deviation exceeds a preset angle threshold and the consistency between the radar equipment and the optoelectronic equipment is good, it is determined that the wireless spectrum detection and direction finding equipment has received deception interference or has malfunctioned, and the fusion weight factor of the wireless spectrum detection and direction finding equipment is set to a preset minimum value. The fusion weight factor is used in the Kalman filter update process. When using the observation data of the wireless spectrum detection and direction finding equipment, the observation noise covariance matrix of the wireless spectrum detection and direction finding equipment is divided by the fusion weight factor to obtain the adjusted observation noise covariance matrix, which is then substituted into the Kalman gain calculation formula to reduce the impact of equipment observations on state estimation updates. If the angle deviation recovers to within the preset angle threshold for M consecutive cycles, the value of the fusion weight factor is restored to the calibration value with a step size of 0.1 per cycle; where M is a positive integer greater than 1.

9. The UAV jamming method based on multi-source information fusion of key defense locations according to claim 1, characterized in that, The step of generating targeted interference parameters by matching a pre-set interference strategy library based on the target type and its confidence level includes: Construct an interference strategy library, which includes at least the following fields: target type, confidence threshold, preferred interference method, preferred interference parameters, alternative interference methods, alternative interference parameters, and switching conditions; wherein, the interference parameters include at least the frequency, power, modulation method, transmitted waveform, and duration of the interference signal; When the confidence level is greater than or equal to the preset confidence threshold, the corresponding preferred interference method and preferred interference parameters are directly matched according to the target type; otherwise, wideband frequency sweep interference is adopted as the default strategy, and the frequency sweep range covers 800MHz to 6GHz, in 10MHz increments. During the interference process, if the effectiveness score of the current interference method is lower than the preset score threshold for a period of time exceeding the preset duration, the method will be switched from the preferred interference method to the alternative interference method. The effectiveness score of the interference method is calculated based on the rate of change of the distance, speed and altitude of the target UAV relative to the defense center.

10. The UAV jamming method based on multi-source information fusion of key defense locations according to claim 1, characterized in that, The process of predicting the target position after mechanical delay of the turntable based on the optimal state estimation, calculating the required horizontal azimuth and pitch angles of the turntable, controlling the turntable rotation, and transmitting interference signals includes: Obtain the target position and target velocity from the optimal state estimate at the current moment; The target position is predicted and compensated based on the mechanical delay time of the turntable: The predicted and compensated target location is obtained; where, This represents the target velocity vector at the current time k. This represents the target position in the optimal state estimate at time k. The target acceleration estimate at the current time k is obtained by the state difference of the current master mode filter, and τ represents the mechanical delay time of the turntable; Transform the predicted and compensated target position into the station center coordinate system with the fixed point of the turntable as the origin, and calculate the relative coordinates and horizontal distance. Calculate the horizontal azimuth and pitch angles of the target UAV; wherein, the horizontal azimuth is the angle rotated clockwise from due north to the target direction, and is calculated using the arctangent function based on the relative coordinates; the pitch angle is the angle positive upwards with the horizontal plane as zero degrees, and is calculated using the arctangent function based on the ratio of the vertical coordinate to the horizontal distance; The calculated horizontal azimuth and pitch angles are sent to the turntable drive unit via the communication interface to control the turntable to rotate to the specified angle. Once the turntable has rotated to its position, an interference signal is transmitted according to the matched interference parameters. When deploying multiple turntables, the turntable closest to the target and with the smallest absolute pitch angle is selected to perform interference. At the same time, the time when the target will leave is predicted based on the coverage of the current turntable. When the remaining coverage time is less than one second, the optimal state estimate and predicted trajectory of the current target are sent to the adjacent turntables, so that the adjacent turntables can rotate to the predicted handover point in advance, thus achieving seamless handover of the interference task.

Images