A smart identification and counter-jamming system for low-altitude aircraft
By fusing radar and radio detection data and using a multi-dimensional feature recognition model, combined with dynamically adjusted judgment thresholds, the problem of identifying and countering low-altitude aircraft in complex environments has been solved, achieving efficient and reliable UAV detection and countermeasures.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG RUITONG ELECTRONIC TECH CO LTD
- Filing Date
- 2026-02-24
- Publication Date
- 2026-05-26
AI Technical Summary
Existing low-altitude aircraft detection and countermeasure systems rely on a single sensor source of information, which makes it impossible to achieve reliable target identification and threat assessment in complex environments. This results in low confidence in identification and judgment and a lack of accurate basis for countermeasure decisions.
Employing radar detection, radio detection, data fusion, and data processing modules, the system identifies drones through multi-dimensional comprehensive feature recognition, combines deep neural networks and temporal behavior validators for target identification, and introduces dynamically adjusted judgment thresholds and multi-source evidence fusion decision-making to achieve accurate identification and countermeasures.
It improves the detection rate in complex environments, reduces the false alarm rate and false alarm rate, achieves the accuracy and environmental adaptability of countermeasures, and enhances the success rate of countermeasures.
Smart Images
Figure CN121711062B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of low-altitude security, specifically to an intelligent identification and counter-interference system for low-altitude aircraft. Background Technology
[0002] With the rapid development of the low-altitude economy and the gradual liberalization of airspace management policies, the application of drones in logistics, inspection, surveying, and entertainment has experienced explosive growth. However, their convenience and widespread use have also brought increasingly serious low-altitude security problems. Unauthorized or maliciously used drones can pose threats of reconnaissance, harassment, and even destruction to key infrastructure, public activity venues, and confidential units. Traditional security methods are insufficient to deal with the intrusion of such "low, slow, and small" targets. Therefore, developing a low-altitude aircraft control system capable of accurate identification, rapid response, and intelligent intervention has become an urgent technological need.
[0003] Currently, mainstream low-altitude aircraft detection and countermeasure systems suffer from a fundamental flaw: they rely on a single sensor source, making reliable target identification and threat assessment impossible in complex environments. Whether using radar, radio detection, or optoelectronic devices alone, only fragments of target attributes, such as location, spectrum, or imagery, can be acquired. Radar struggles to distinguish between drones and birds, radio detection fails when the target signal is silent, and optoelectronic methods are limited by weather and lighting conditions. This lack of perception dimension prevents the system from building a complete cognitive map of the target, resulting in low confidence in subsequent identification and judgment, and a lack of precise basis for countermeasure decisions. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent identification and counter-interference system for low-altitude aircraft, thereby solving the above-mentioned technical problems.
[0005] The objective of this invention can be achieved through the following technical solutions:
[0006] A smart identification and counter-jamming system for low-altitude aircraft includes:
[0007] The radar detection module is used to transmit electromagnetic wave signals into the monitored airspace and receive the echo signals reflected by the target in order to obtain the target's distance, speed, azimuth and micro-motion information.
[0008] The radio detection module is used to scan and receive radio spectrum signals in the monitored airspace, and to intercept and analyze radio signals that conform to the parameter characteristics of UAV remote control, image transmission or navigation signals.
[0009] The data fusion module is used to receive and fuse target status information from the radar detection module and signal characteristic information from the radio detection module.
[0010] The data processing module, connected to the data fusion module, includes a feature extraction unit for extracting multi-dimensional comprehensive features, and an identification unit for determining whether a target is a drone based on the multi-dimensional comprehensive features.
[0011] The early warning and countermeasure module is connected to the data processing module and is used to generate early warning information based on the identification results of the identification unit and determine whether to activate interference countermeasures.
[0012] As a further technical solution, the multi-dimensional integrated features include: target micro-Doppler spectrum features, radar cross-section fluctuation features, radio signal fingerprint features, and motion trajectory features; the identification unit inputs the multi-dimensional integrated features into a pre-trained UAV identification model and outputs whether the target is a UAV. If the target is a UAV, it outputs the UAV's model, threat level, and status information.
[0013] As a further technical solution, the feature extraction unit performs the following steps to process the multidimensional comprehensive features:
[0014] The original echo sequence of the target provided by the radar detection module is obtained, and the micro-Doppler spectrum features of the target are extracted by short-time Fourier transform or wavelet transform. The micro-Doppler spectrum features are then reduced in dimensionality using principal component analysis, and the first N principal component coefficients representing the periodic motion of the rotor or propeller are used as the first feature vector.
[0015] The radar cross-section sequence of the target within the observation time window is statistically analyzed, and the mean, variance, skewness, and kurtosis are calculated. The mean, variance, skewness, and kurtosis are then used as the characteristic vector of radar cross-section fluctuation.
[0016] For signals intercepted by the radio detection module, the statistical moments of instantaneous amplitude, instantaneous frequency, and instantaneous phase are calculated, and cyclic spectral density analysis is performed to extract feature ridges at a specified cyclic frequency, which together constitute a radio signal fingerprint feature vector; the feature ridges contain intensity and shape structure information;
[0017] For the target trajectory point sequence generated by the data fusion module, the tangential velocity, normal acceleration, trajectory curvature and information entropy of the trajectory point distribution in three-dimensional spatial coordinates are calculated respectively to form the motion trajectory feature vector;
[0018] After normalizing the first feature vector, radar cross-section undulation feature vector, radio signal fingerprint feature vector, and motion trajectory feature vector, they are aligned and concatenated in time sequence to generate a unified multidimensional feature tensor for input recognition unit.
[0019] As a further technical solution, the drone identification model specifically includes:
[0020] A primary classifier based on a deep neural network or gradient boosting decision tree is used to perform preliminary calculations on the unified multidimensional feature tensor and output the primary probability and preliminary model classification of the target being a UAV.
[0021] A temporal behavior validator based on temporal convolutional networks or long short-term memory networks receives the continuous time series feature tensor corresponding to the target identified as a drone by the primary classifier, analyzes the temporal consistency between micro-Doppler features and motion trajectory features, and outputs the temporal verification confidence.
[0022] The multi-source evidence fusion decision module performs decision-level fusion using a Bayesian fusion method based on primary probability, time-series verification confidence, and the matching degree between radio signal fingerprint features and preset model templates. Only when the confidence after fusion exceeds a dynamically adjusted judgment threshold will the target be finally determined to be a drone and the model, threat level, and status information will be output.
[0023] As a further technical solution, the early warning and countermeasure module also includes a dynamic verification unit, specifically including:
[0024] The real-time comprehensive threat index is calculated based on the drone model, threat level, and real-time location output by the identification unit and the degree of overlap with sensitive areas in the preset electronic map.
[0025] When the real-time comprehensive threat index exceeds the first threshold, the identification unit is allowed to independently identify the same target in the subsequent M consecutive detection cycles; if the proportion of the number of times the target is identified as a drone in the M identification results exceeds the preset proportion, and the average confidence after fusion is higher than the review threshold dynamically adjusted according to the false alarm rate of the current environment, then the target threat is confirmed.
[0026] Based on the confirmed threat level and drone model, the countermeasure strategy library is queried to generate an initial countermeasure command that includes the specific countermeasure technology type, expected effect parameters, and priority.
[0027] As a further technical solution, the early warning and countermeasure module also includes a countermeasure execution unit, specifically including:
[0028] Execute the countermeasures corresponding to the initial countermeasure command and start an evaluation window of a preset duration;
[0029] Within the evaluation window, the rate of change of distance of the target relative to the core sensitive region, the intensity attenuation of the characteristic ridge line of the target-associated signal, and the structural distortion are continuously acquired and calculated.
[0030] If the distance change rate does not turn from negative to positive, and the intensity attenuation and structural distortion of the characteristic ridge line do not reach the preset threshold, then it is determined to be a primary countermeasure failure.
[0031] When the primary countermeasure fails, an alternative countermeasure action is selected and executed from a library of alternative countermeasure actions based on different technical principles.
[0032] As a further technical solution, the specific working method of the data fusion module is as follows:
[0033] Within a single fusion cycle, for the same target, the target spatial state information and corresponding body confidence provided by the radar detection module are acquired. The body confidence is calculated based on the signal-to-noise ratio of the radar echo and the stability of the target radar cross-section.
[0034] The target radio signal matching information and corresponding link confidence provided by the radio detection module are obtained. The link confidence is calculated based on the salience of the extracted signal feature ridge and the persistence of the signal.
[0035] Calculate the scene adaptation factor for radar data and the scene adaptation factor for radio data respectively;
[0036] Among them, the scene adaptation factor of radar data is obtained by multiplying the normalized value calculated based on the radar echo signal-to-noise ratio with the proportion of the target micro-motion phase change to the theoretical maximum value.
[0037] The scene adaptation factor for radio data is obtained by multiplying the ratio of the power of the extracted feature ridges to the total power with the proportion of the signal coherence time to the observation window duration.
[0038] The final fusion weights are determined by normalizing the product of the ontology confidence, the link confidence, and the corresponding scene adaptation factor; and the radar target state information and radio signal matching information are weighted and fused based on these weights.
[0039] As a further technical solution, the dynamic adjustment judgment threshold is equal to the basic threshold plus the sum of the system cognitive adjustment amount and the environmental disturbance adjustment amount;
[0040] The system cognitive adjustment quantity is the product of the system cognitive bias and the system cognitive factor. The system cognitive bias is obtained by calculating the average difference between the predicted probability output by the identification unit and the final verification confirmation result within the historical period, and is used to quantify the stability of the system's own judgment.
[0041] The environmental interference adjustment amount is the product of the environmental interference index and the environmental interference factor. The environmental interference index is obtained by comprehensively calculating the spectrum occupancy ratio of the current non-cooperative signal and the frequency and duration characteristics of the transient interference signal.
[0042] As a further technical solution, the real-time comprehensive threat index is obtained through the following method:
[0043] The static baseline threat value and the dynamic evolution threat value of the target are calculated separately, and the maximum value of the two is taken as the final real-time comprehensive threat index.
[0044] The static baseline threat value is the product of the inherent threat level corresponding to the target UAV model, the preset sensitivity coefficient of the target's location area, and the enhancement coefficient for the current time period.
[0045] The dynamic evolution threat value is calculated by multiplying the target's radial normalized velocity by an exponential decay factor based on normalized distance, and adding an anomaly quantification factor for the flight trajectory. The radial normalized velocity is the ratio of the target's approach velocity along the line connecting to the core area to the target model's maximum nominal velocity. The normalized distance is the actual distance between the target's current position and the core area divided by the preset critical defense distance of that area. The trajectory anomaly quantification factor is the difference between the information entropy of the target's flight trajectory in the previous cycle and the information entropy of the normal trajectory of a similar UAV.
[0046] The beneficial effects of this invention are:
[0047] (1) This invention uses data fusion of radar detection and radio detection to comprehensively utilize the target's micro-Doppler spectrum, radar cross-section fluctuations, radio signal fingerprints and motion trajectory features for identification, overcoming the limitations of a single sensor source; it adopts a UAV identification model that includes primary classification, time-series verification and multi-source evidence fusion, and introduces dynamically adjusted judgment thresholds, which improves the detection rate of UAVs in complex environments, while significantly reducing false alarm rate and false alarm rate;
[0048] (2) This invention calculates a real-time comprehensive threat index that integrates static attributes and dynamic behavior, and performs dynamic verification and confirmation of high-threat targets, thereby achieving the accuracy of countermeasure initiation. Furthermore, the countermeasure execution process includes an effect evaluation window, which can judge the effectiveness of countermeasures in real time based on the target's distance changes and signal attenuation. When the primary countermeasure fails, it can switch to alternative countermeasures based on different principles in a timely manner, thereby improving the success rate and environmental adaptability of countermeasures. Attached Figure Description
[0049] The invention will now be further described with reference to the accompanying drawings.
[0050] Figure 1 This is a logical structure diagram of the present invention. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] Please see Figure 1 As shown, the present invention is an intelligent identification and counter-jamming system for low-altitude aircraft, comprising:
[0053] The radar detection module is used to transmit electromagnetic wave signals into the monitored airspace and receive the echo signals reflected by the target in order to obtain the target's distance, speed, azimuth and micro-motion information; thus providing the system with the perception dimension of the target's spatial motion state and micro-motion characteristics.
[0054] The radio detection module is used to scan and receive radio spectrum signals in the monitored airspace, and to intercept and analyze radio signals that conform to the parameter characteristics of UAV remote control, image transmission, or navigation signals. The parameter characteristics conforming to UAV signals refer to a frequency range of 2.4GHz±50MHz and 5.8GHz±50MHz, a modulation method of FSK / GFSK, and a bandwidth ≤20MHz. This provides the system with an independent sensing dimension—the radio signal fingerprint of the target—complementing radar detection.
[0055] The data fusion module receives and fuses target status information from the radar detection module and signal characteristic information from the radio detection module; its specific operation is as follows:
[0056] Within a single fusion cycle, for the same target, the target spatial state information and corresponding body confidence provided by the radar detection module are acquired. The body confidence is calculated based on the signal-to-noise ratio of the radar echo and the stability of the target radar cross-section.
[0057] Specifically:
[0058] The signal-to-noise ratio of the radar echo is normalized and used as the signal-to-noise ratio confidence component.
[0059] The stability of the target radar cross-section is quantified by the variance of its radar cross-section sequence within the observation time window. This sequence is also used to generate the radar cross-section fluctuation feature vector as a stability confidence component.
[0060] The ontology confidence score is a weighted sum or product of the signal-to-noise ratio confidence component and the stability confidence component, both of which have been normalized. This calculation method quantifies the reliability of radar data, providing a basis for weighted fusion.
[0061] The target radio signal matching information and corresponding link confidence provided by the radio detection module are obtained. The link confidence is calculated based on the salience of the extracted signal feature ridge and the persistence of the signal.
[0062] Specifically:
[0063] Significance is quantified by the ratio of the power of the extracted feature ridges to the total power, and this ratio is also used to calculate the scene adaptation factor of the radio data;
[0064] Persistence is quantified by the ratio of signal coherence time to the duration of the observation window, and this ratio is also used to calculate the scene adaptation factor of radio data;
[0065] Link confidence is a weighted sum or product of saliency and persistence components, both of which have been normalized. This calculation method quantifies the reliability of radio signals, providing another basis for weighted fusion.
[0066] Calculate the scene adaptation factor for radar data and the scene adaptation factor for radio data respectively;
[0067] Among them, the scene adaptation factor of radar data is obtained by multiplying the radar echo signal-to-noise ratio obtained by linearly normalizing it to the [0,1] interval with the ratio of the target micro-motion phase change to the theoretical maximum value; this factor dynamically reflects the radar's perception effectiveness under the current environment and target characteristics.
[0068] The scene adaptation factor for radio data is obtained by multiplying the ratio of the power of the extracted feature ridge to the total power and the proportion of the signal coherence time to the observation window duration; the total power is the sum of the power of the feature ridge and the power of the ambient noise; this factor dynamically reflects the sensing effectiveness of radio in the current electromagnetic environment.
[0069] The final fusion weights are determined by normalizing the products of ontology confidence, link confidence, and the corresponding scene adaptation factor. These weights are then used to weight and fuse radar target state information with radio signal matching information. This fusion mechanism adaptively adjusts the contribution of different sensors based on their real-time data quality, significantly improving the overall robustness and data reliability of the system's perception in complex environments.
[0070] The dynamic adjustment judgment threshold is equal to the basic threshold plus the sum of the system cognitive adjustment amount and the environmental disturbance adjustment amount;
[0071] The system cognitive adjustment quantity is the product of the system cognitive bias and the system cognitive factor. The system cognitive bias is obtained by calculating the average difference between the predicted probability output by the identification unit and the final verification result within the historical period, and is used to quantify the stability of the system's own judgment. This design enables the system to self-assess its identification accuracy.
[0072] The environmental interference adjustment amount is the product of the environmental interference index and the environmental interference factor. The environmental interference index is obtained by comprehensively calculating the spectral occupancy ratio of the current non-cooperative signal and the frequency and duration characteristics of the transient interference signal. This design enables the system to perceive the complexity of the environment. The combined effect of these two factors allows the judgment threshold to be dynamically adjusted according to changes in system state and environment, thereby effectively suppressing false alarms while ensuring a high detection rate.
[0073] The data processing module includes a feature extraction unit for extracting multidimensional comprehensive features, which include: target micro-Doppler spectrum features, radar cross-section undulation features, radio signal fingerprint features, and motion trajectory features. By extracting and fusing these four types of features, multidimensional and complementary descriptive information is provided for target recognition, overcoming the limitations of single feature recognition capabilities.
[0074] The feature extraction unit performs the following steps to process the multidimensional integrated features:
[0075] The system acquires the original echo sequence of the target provided by the radar detection module. Micro-Doppler spectral features of the target are extracted using short-time Fourier transform or wavelet transform. Principal component analysis is then used to reduce the dimensionality of these micro-Doppler spectral features, and the first N principal component coefficients representing the periodic motion of the rotor or propeller are used as the first feature vector. In these first N principal component coefficients, N is the smallest integer satisfying a cumulative variance contribution rate of ≥90% for the first N principal components, typically between 5 and 10. This step effectively extracts and condenses the unique micro-motion fingerprint information of the UAV rotor / propeller.
[0076] The radar cross-section sequence of the target within the observation time window is statistically analyzed, and the mean, variance, skewness, and kurtosis are calculated. The mean, variance, skewness, and kurtosis are then used as the characteristic vector of radar cross-section fluctuation. This step quantifies the statistical regularity of the target's scattering characteristics, which helps to distinguish aircraft with different geometric structures.
[0077] For signals intercepted by the radio detection module, the statistical moments of instantaneous amplitude, instantaneous frequency, and instantaneous phase are calculated, and cyclic spectral density analysis is performed to extract feature ridges at specified cyclic frequencies, which together constitute the radio signal fingerprint feature vector; the feature ridges contain intensity and shape structure information; this step deeply mines the modulation details and periodic characteristics of UAV remote control and image transmission signals, forming a signal identity that is difficult to imitate.
[0078] As one implementation method, the characteristic ridge intensity was detected by a spectrum analyzer. The measured ridge intensity of the DJI Mavic 3 drone was 65dB, and that of the self-made racing drone was 58dB. The slope variance was calculated using MATLAB software and was set to 0.08. The number of peaks was 4, all of which met the quantization requirements. The test method adopted a standard procedure of signal acquisition, spectrum decomposition, ridge fitting, and parameter calculation, which can be directly referred to and implemented by those skilled in the art.
[0079] For the target trajectory point sequence generated by the data fusion module, the tangential velocity, normal acceleration, trajectory curvature, and information entropy of the trajectory point distribution in three-dimensional spatial coordinates are calculated to form a motion trajectory feature vector. This step characterizes the target's maneuvering mode and intention from the perspective of spatial kinematics, providing behavioral basis for identification.
[0080] After normalizing the first feature vector, radar cross-section undulation feature vector, radio signal fingerprint feature vector, and motion trajectory feature vector, they are aligned and concatenated in time sequence to generate a unified multidimensional feature tensor for input to the recognition unit. This step achieves standardization and spatiotemporal alignment of heterogeneous feature vectors, providing a unified and complete input for subsequent intelligent recognition models.
[0081] The identification unit inputs the multidimensional integrated features into a pre-trained UAV identification model and outputs whether the target is a UAV. If the target is a UAV, it outputs the UAV's model, threat level, and status information.
[0082] The training data for the UAV recognition model includes real flight data from over 1,000 different UAV models, including distance, speed, and signal characteristics; and over 500 sets of simulated electromagnetic interference data. The data annotations are confirmed manually in conjunction with raw radar and radio data. The model is trained using cross-validation with 1,000 iterations, and the convergence condition is a loss function ≤ 0.01. Through large-scale, high-quality data training, the recognition model is ensured to have strong generalization ability and high accuracy.
[0083] The drone identification model specifically includes:
[0084] A primary classifier based on a deep neural network or gradient boosting decision tree is used to perform preliminary calculations on the unified multidimensional feature tensor and output the primary probability and preliminary model classification of the target as a UAV; thus achieving a fast and preliminary determination of the target's identity.
[0085] If a deep neural network (DNN) is used: it contains 3 hidden layers, with 256 neurons in the first layer, 128 in the second layer, and 64 in the third layer. The activation function is ReLU, and the output layer uses the Softmax function. The output dimension is equal to the number of drone model categories plus 1 (1 represents non-drone categories).
[0086] If Gradient Boosting Decision Tree (GBDT) is used: the number of decision trees is 100-150, the learning rate is 0.05-0.1, the maximum depth of each tree is ≤8, and the loss function is cross-entropy; the above parameter range (such as tree depth limit) is intended to balance model complexity and generalization ability and prevent overfitting, and is determined through grid search and cross-validation.
[0087] The training objective of the primary classifier is to minimize the classification error between UAVs and non-UAVs, as well as the multi-class classification error. The training data uses a combination of a unified multidimensional feature tensor and manually labeled data. The optimization algorithm is Adam (DNN) or gradient descent (GBDT).
[0088] A temporal behavior validator based on temporal convolutional networks or long short-term memory networks receives the continuous time-series feature tensor corresponding to the target identified as a drone by the primary classifier, analyzes the temporal consistency between micro-Doppler features and motion trajectory features, and outputs the temporal verification confidence score. This design introduces temporal dimension verification, which can effectively filter out single-frame misjudgments caused by instantaneous interference or feature coincidence, and significantly improve the reliability of recognition.
[0089] If a Temporal Convolutional Network (TCN) is used: the kernel size is 3, the number of hidden layers is 2, the dilation coefficients are 1 and 2 respectively, and the output dimension is 1;
[0090] If a Long Short-Term Memory (LSTM) network is used: it contains 2 hidden layers, each with 64 units, a dropout ratio of 0.2, and an output dimension of 1;
[0091] The confidence score for time series validation is equal to the normalized output value (range 0 to 1). When the confidence score is ≥ 0.8, the time series features are considered consistent. This threshold of 0.8 and the time series window length of 5 periods were determined after performance evaluation of historical validation datasets (e.g., maximizing the F1 score). This threshold setting balances rigor with practicality.
[0092] The input to the temporal behavior validator is a multidimensional feature tensor for 5 consecutive detection cycles, i.e., temporal window length = 5. The core verification logic is whether the periodicity of the micro-Doppler spectrum and the smoothness of the motion trajectory match the motion law of the UAV.
[0093] The multi-source evidence fusion decision module performs decision-level fusion using a Bayesian fusion method, based on primary probability, temporal verification confidence, and the matching degree between radio signal fingerprint features and preset model templates. Only when the fused confidence exceeds a dynamically adjusted judgment threshold is the target ultimately determined to be a drone, and its model, threat level, and status information are output. This design, by fusing evidence from different stages and dimensions and setting a dynamic judgment threshold, achieves a high-confidence final decision, significantly reducing the possibility of false alarms and missed detections.
[0094] The preset model template is a feature set for known drone models, including:
[0095] Core components: radio signal fingerprint feature vector (statistical moments of instantaneous amplitude / frequency / phase, feature ridge parameters), principal component coefficient range of micro-Doppler spectrum features, and typical motion trajectory features (tangential velocity range, trajectory curvature range).
[0096] Matching degree calculation: The preset model template vector is the historical average of the feature vector of the model; when calculating the matching degree, the feature vector of the target to be identified and the template vector are used to calculate the cosine similarity, and the quantization range is 0 to 1.
[0097] The preset model templates are built based on a large amount of measured data from known UAV models: more than 50 flight data points are collected for each model, features are extracted, and the mean ± 3 times the standard deviation is used as the template range. The template library supports online updates, and automatic iterative optimization is performed after new model data is added. This template library establishment and update mechanism enables the system to continuously learn and adapt to new UAV models.
[0098] The early warning and countermeasure module is used to generate early warning information based on the identification results of the identification unit and to determine whether to activate interference countermeasures; it also includes a dynamic verification unit, specifically comprising:
[0099] Based on the drone model, threat level, and real-time location output by the identification unit and the degree of overlap with sensitive areas in the preset electronic map, a real-time comprehensive threat index is calculated. This step combines the inherent attributes, dynamic behavior, and geographical protection requirements of the target to generate a quantified threat metric.
[0100] When the real-time comprehensive threat index exceeds the first threshold, the identification unit is allowed to independently identify the same target in the subsequent M consecutive detection cycles. This design avoids blindly following the results of a single identification and introduces redundant verification of the decision. The first threshold can be preset between 0.5 and 0.8 according to the coreness of the protected area, or obtained by reverse calibration through historical intrusion event data.
[0101] If, in M identification results, the proportion of times an object is identified as a drone exceeds a preset proportion, and the average confidence level after fusion is higher than the review threshold dynamically adjusted based on the current environmental false alarm rate, then the target threat is confirmed. Environmental false alarm rate = number of non-targets misidentified as drones in the past 100 detection cycles ÷ total number of detected targets. This review mechanism, through the dual constraints of statistical judgment and confidence threshold, effectively suppresses the interference of environmental false alarms on decision-making while improving the reliability of threat confirmation.
[0102] In M consecutive detection cycles, M is 3-5; the preset ratio is ≥80%; the review threshold = basic review threshold × (1 + current environmental false alarm rate), and the basic review threshold is preset to 0.85;
[0103] Based on the confirmed threat level and drone model, the system queries the countermeasure strategy library to generate initial countermeasure commands containing specific countermeasure technology types, expected action parameters, and priorities. The countermeasure strategy library is categorized in three dimensions: drone model, threat level, and flight status. For example, low-threat civilian drones (such as the DJI Mavic) are targeted with protocol-level spoofing injection, while medium-to-high-threat drones (such as industrial surveying drones) are targeted with directional beamforming jamming. Countermeasure parameter matching rules: beam angle = target azimuth ± 5°, jamming power = target signal power × 10-20 times. This strategy library achieves differentiated, precise, and parameterized countermeasures, avoiding the inefficiency or collateral damage caused by a one-size-fits-all approach.
[0104] The early warning and countermeasure module also includes a countermeasure execution unit, specifically comprising:
[0105] The countermeasures corresponding to the initial countermeasure command are executed, and an evaluation window of a preset duration is initiated; this design couples the execution and evaluation processes, forming a closed-loop control basis for countermeasures.
[0106] Within the evaluation window, the distance change rate of the target relative to the core sensitive region, the intensity attenuation of the characteristic ridge line of the target-associated signal, and the structural distortion are continuously acquired and calculated; the countermeasure effect is quantified from two dimensions: spatial behavior and signal state.
[0107] If the distance change rate does not turn from negative to positive, and the intensity attenuation and structural distortion of the characteristic ridge line do not reach the preset threshold, it is determined that the primary countermeasure has failed. The preset threshold for the countermeasure effect is: intensity attenuation ≥ 30dB, structural distortion ≥ 60%. By setting clear and quantifiable effect judgment criteria, an objective assessment of the effectiveness of the countermeasure is achieved.
[0108] The structural distortion is quantified by calculating the relative rate of change of the shape parameters (such as slope variance and number of peaks) of the feature ridge after countermeasure with the baseline value before countermeasure. The calculation formula is: (baseline value - current value) / baseline value × 100%.
[0109] When the initial countermeasure fails, an option is selected and executed from a library of alternative countermeasures based on different technical principles. This design gives the system the ability to adaptively upgrade in the event of an initial countermeasure failure, thereby improving the overall countermeasure success rate.
[0110] As one implementation method, in the event of multi-target intrusion, targets are sorted in descending order according to the real-time comprehensive threat index, and countermeasures are prioritized against the top K targets, where K is the maximum number of countermeasure targets the system can handle simultaneously, typically ranging from 1 to 3. In the event of radar or radio module failure, the system automatically switches to single-module operation mode; for example, in the event of radar module failure, identification is based on radio fingerprints and motion trajectory recognition, while in the event of radio module failure, identification is based on radar multi-dimensional features. When the target signal is interrupted, the feature data from the previous three cycles is retained, the trajectory is continuously tracked, and the identification is re-fused after the signal is restored. These strategies enhance the system's survivability and continuous operation capability in complex scenarios such as multi-target intrusion, module failure, and target concealment.
[0111] The alternative actions include, but are not limited to:
[0112] Switching from continuous wave interference to intermittent protocol-level deception injection synchronized with the target signal protocol;
[0113] Switching from omnidirectional suppression to directional beamforming jamming based on the real-time location of the target;
[0114] The system sends a forged command to the target, conforming to its communication protocol and containing pre-set safe landing point coordinates. A library of alternative actions provides countermeasures based on various technical principles, ensuring that if one method fails, other mechanisms remain available, thus improving the robustness and adaptability of the countermeasure system.
[0115] The dynamically adjusted judgment threshold is determined by: Calculated;
[0116] in This is a baseline threshold determined based on historical data; This is the average of the absolute differences between the prediction probability of whether the target is a drone by the identification unit and the final verification result over the past N periods; , The adjustment coefficient can be set as follows: =0.3、 =0.2, obtained through offline optimization of the gradient descent algorithm based on historical data from the past 3 months with the goal of minimizing the overall misclassification rate of the system;
[0117] Environmental disturbance index; Where B(t) is the current spectrum width occupied by non-cooperative signals, Ball is the total detection bandwidth, Ns(t) is the number of transient non-UAV signal pulses occurring per unit time, and Tavg is the average duration.
[0118] The real-time comprehensive threat index is obtained through the following methods:
[0119] The static baseline threat value and the dynamic evolution threat value of the target are calculated separately, and the maximum value of the two is taken as the final real-time comprehensive threat index.
[0120] The static baseline threat value is the product of the inherent threat level corresponding to the target UAV model, the preset sensitivity coefficient of the target's location area, and the enhancement coefficient for the current time period. The enhancement coefficient for the current time period is set according to the frequency of historical threat events, for example: 1.5 for sensitive periods (00:00-06:00, major event periods) and 1.0 for regular periods.
[0121] The dynamic evolutionary threat value = (radial normalized velocity × exponential decay factor based on normalized distance) + trajectory anomaly quantification factor; where the exponential decay factor is... ; For normalized distance, The attenuation coefficient is obtained by fitting the relationship curve between threat level and distance in historical events, and a typical value is 0.5. An attenuation coefficient of 0.5 causes the threat value to decrease exponentially as the target moves away from the core area; that is, the closer the target is (d is 0), the attenuation factor is 1, and the dynamic evolution threat value is not affected by distance attenuation; the farther the target is (d is 1), the attenuation factor is 0.606, and the dynamic evolution threat value is moderately reduced. This calculation method makes the dynamic evolution threat value more sensitive to the approach behavior of nearby targets.
[0122] The radially normalized velocity is the ratio of the target's approach velocity along the line connecting to the core region to the target model's maximum nominal velocity;
[0123] The normalized distance is the actual distance between the target's current location and the core area divided by the preset critical defense distance of that area;
[0124] The trajectory anomaly quantification factor is the difference between the information entropy of the target's previous flight trajectory and the information entropy of the normal trajectory of a similar UAV.
[0125] The normal trajectory information entropy of the same type of UAV is the baseline value of the trajectory information entropy of this type of UAV in a non-threat state, obtained by learning from historical data.
[0126] It should be noted that the calculation formulas and all parameters involved in the calculations in this invention have been dimensionless beforehand. The process of dimensionless processing is well known in the industry and will not be described here.
[0127] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.
Claims
1. An intelligent identification and counter-interference system for low-altitude flying vehicles, characterized in that, include: The radar detection module is used to transmit electromagnetic wave signals into the monitored airspace and receive the echo signals reflected by the target in order to obtain the target's distance, speed, azimuth and micro-motion information. The radio detection module is used to scan and receive radio spectrum signals in the monitored airspace, and to intercept and analyze radio signals that conform to the parameter characteristics of UAV remote control, image transmission or navigation signals. The data fusion module is used to receive and fuse target status information from the radar detection module and signal characteristic information from the radio detection module. The data processing module, connected to the data fusion module, includes a feature extraction unit for extracting multi-dimensional comprehensive features, and an identification unit for determining whether a target is a drone based on the multi-dimensional comprehensive features. The early warning and countermeasure module is connected to the data processing module and is used to generate early warning information based on the identification result of the identification unit and determine whether to activate interference countermeasures. The early warning and countermeasure module also includes a dynamic verification unit, specifically comprising: The real-time comprehensive threat index is calculated based on the drone model, threat level, and real-time location output by the identification unit and the degree of overlap with sensitive areas in the preset electronic map. When the real-time comprehensive threat index exceeds the first threshold, the identification unit is allowed to independently identify the same target in the subsequent M consecutive detection cycles; if the proportion of the number of times the target is identified as a drone in the M identification results exceeds the preset proportion, and the average confidence after fusion is higher than the review threshold dynamically adjusted according to the false alarm rate of the current environment, then the target threat is confirmed. Based on the confirmed threat level and drone model, the countermeasure strategy library is queried to generate an initial countermeasure command that includes the specific countermeasure technology type, expected effect parameters, and priority.
2. The intelligent identification and countermeasure jamming system for low altitude flying objects according to claim 1, characterized in that, The multidimensional integrated features include: target micro-Doppler spectrum features, radar cross-section fluctuation features, radio signal fingerprint features, and motion trajectory features; The identification unit inputs the multidimensional integrated features into a pre-trained UAV identification model and outputs whether the target is a UAV. If the target is a UAV, it outputs the UAV's model, threat level, and status information.
3. The intelligent identification and counter-ECM system for low-flying aerial vehicles according to claim 2, characterized in that, The feature extraction unit performs the following steps to process the multidimensional integrated features: The original echo sequence of the target provided by the radar detection module is obtained, and the micro-Doppler spectrum features of the target are extracted by short-time Fourier transform or wavelet transform. The micro-Doppler spectrum features are then reduced in dimensionality using principal component analysis, and the first N principal component coefficients representing the periodic motion of the rotor or propeller are used as the first feature vector. The radar cross-section sequence of the target within the observation time window is statistically analyzed, and the mean, variance, skewness, and kurtosis are calculated. The mean, variance, skewness, and kurtosis are then used as the radar cross-section fluctuation feature vector. For the signal intercepted by the radio detection module, the statistical moments of instantaneous amplitude, instantaneous frequency, and instantaneous phase are calculated, and cyclic spectral density analysis is performed to extract feature ridges at a specified cyclic frequency, which together constitute a radio signal fingerprint feature vector; the feature ridges contain intensity and shape structure information; For the target trajectory point sequence generated by the data fusion module, the tangential velocity, normal acceleration, trajectory curvature, and information entropy of the trajectory point distribution in three-dimensional spatial coordinates are calculated respectively to form a motion trajectory feature vector; After normalizing the first feature vector, the radar cross-section undulation feature vector, the radio signal fingerprint feature vector, and the motion trajectory feature vector, they are aligned and concatenated in time sequence to generate a unified multidimensional feature tensor for input to the identification unit.
4. The intelligent identification and counter-ECM system for low-flying aerial vehicles according to claim 3, characterized in that, The drone identification model specifically includes: A primary classifier based on a deep neural network or gradient boosting decision tree is used to perform preliminary calculations on the unified multidimensional feature tensor and output the primary probability and preliminary model classification of the target being a UAV. A temporal behavior validator based on temporal convolutional networks or long short-term memory networks receives the continuous time series feature tensor corresponding to the target identified as a drone by the primary classifier, analyzes the temporal consistency between micro-Doppler features and motion trajectory features, and outputs the temporal verification confidence. The multi-source evidence fusion decision module performs decision-level fusion using a Bayesian fusion method based on primary probability, time-series verification confidence, and the matching degree between radio signal fingerprint features and preset model templates. Only when the confidence after fusion exceeds a dynamically adjusted judgment threshold will the target be finally determined to be a drone and the model, threat level, and status information will be output.
5. The intelligent identification and countermeasure jamming system for low-flying aerial vehicles according to claim 1, characterized in that, The early warning and countermeasure module also includes a countermeasure execution unit, specifically comprising: Execute the countermeasures corresponding to the initial countermeasure command and start an evaluation window of a preset duration; Within the evaluation window, the rate of change of distance of the target relative to the core sensitive region, the intensity attenuation of the characteristic ridge line of the target-associated signal, and the structural distortion are continuously acquired and calculated. If the distance change rate does not turn from negative to positive, and the intensity attenuation and structural distortion of the characteristic ridge line do not reach the preset threshold, then it is determined to be a primary countermeasure failure. When the primary countermeasure fails, an alternative countermeasure action is selected and executed from a library of alternative countermeasure actions based on different technical principles.
6. The intelligent identification and counter-countermeasure jamming system of low-altitude flying craft according to claim 3, characterized in that, The specific working principle of the data fusion module is as follows: Within a single fusion cycle, for the same target, the target spatial state information and corresponding body confidence provided by the radar detection module are acquired. The body confidence is calculated based on the signal-to-noise ratio of the radar echo and the stability of the target radar cross-section. The target radio signal matching information and corresponding link confidence provided by the radio detection module are obtained. The link confidence is calculated based on the salience of the extracted signal feature ridge and the persistence of the signal. Calculate the scene adaptation factor for radar data and the scene adaptation factor for radio data respectively; Among them, the scene adaptation factor of radar data is obtained by multiplying the normalized value calculated based on the radar echo signal-to-noise ratio with the proportion of the target micro-motion phase change to the theoretical maximum value. The scene adaptation factor for radio data is obtained by multiplying the ratio of the power of the extracted feature ridges to the total power with the proportion of the signal coherence time to the observation window duration. The final fusion weights are determined by normalizing the product of the ontology confidence, the link confidence, and the corresponding scene adaptation factor; and the radar target state information and radio signal matching information are weighted and fused based on these weights.
7. The intelligent identification and counter-jamming system for low-altitude aircraft according to claim 4, characterized in that, The dynamic adjustment judgment threshold is equal to the basic threshold plus the sum of the system cognitive adjustment amount and the environmental disturbance adjustment amount; The system cognitive adjustment quantity is the product of the system cognitive bias and the system cognitive factor. The system cognitive bias is obtained by calculating the average difference between the predicted probability output by the identification unit and the final verification confirmation result within the historical period, and is used to quantify the stability of the system's own judgment. The environmental interference adjustment amount is the product of the environmental interference index and the environmental interference factor. The environmental interference index is obtained by comprehensively calculating the spectrum occupancy ratio of the current non-cooperative signal and the frequency and duration characteristics of the transient interference signal.
8. The intelligent identification and counter-jamming system for low-altitude aircraft according to claim 1, characterized in that, The real-time comprehensive threat index is obtained through the following methods: The static baseline threat value and the dynamic evolution threat value of the target are calculated separately, and the maximum value of the two is taken as the final real-time comprehensive threat index. The static baseline threat value is the product of the inherent threat level corresponding to the target UAV model, the preset sensitivity coefficient of the target's location area, and the enhancement coefficient for the current time period. The dynamic evolution threat value is calculated by multiplying the target's radial normalized velocity by an exponential decay factor based on normalized distance, and adding an anomaly quantification factor for the flight trajectory. The radial normalized velocity is the ratio of the target's approach velocity along the line connecting to the core area to the target model's maximum nominal velocity. The normalized distance is the actual distance between the target's current position and the core area divided by the preset critical defense distance of that area. The trajectory anomaly quantification factor is the difference between the information entropy of the target's flight trajectory in the previous cycle and the information entropy of the normal trajectory of a similar UAV.