Methods and systems for drone detection and control in regional defense

Abstract

This invention relates to a method and system for detecting and controlling unmanned aerial vehicles (UAVs) for regional defense. The method involves dividing the area to be defended into sub-regions, setting differentiated no-fly thresholds and detection response priorities. Secondly, based on the sub-region division, a multi-dimensional collaborative detection network is constructed, integrating radio spectrum monitoring, millimeter-wave radar, electro-optical tracking, and acoustic sensors to collect multi-source data from UAVs in real time. Coordinate matching is used to determine the sub-region to which the UAV belongs and to filter out over-limit target data, extracting target feature sets based on flight speed, altitude, communication protocol type, radar cross-section, and acoustic signature information. A dynamic target classification model is input, dynamically outputting three threat levels: harmless civilian, suspicious reconnaissance, and high-risk attack. Tiered control is implemented by combining sub-region priorities and threat levels, and the control effect is dynamically evaluated based on preset assessment indicators. Detection parameters and control strategies are adaptively adjusted to effectively reduce the risk of intrusion into the core area and enhance the intelligence and adaptability of regional defense.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) management technology, and particularly relates to UAV detection and management methods and systems for regional defense. Background Technology

[0002] With the rapid development of drone technology, the number and application scenarios of consumer and industrial drones have increased significantly. While bringing convenience, this has also led to serious security risks. According to industry statistics, there were over 5,000 drone-related security incidents globally in 2023, with unauthorized intrusions into core sensitive areas accounting for 32% of these incidents. Some incidents even involved malicious intrusions carrying dangerous items, posing a direct threat to regional security. Against this backdrop, how to achieve efficient detection and precise control of intruding drones has become a core issue that urgently needs to be addressed in the security field.

[0003] Current mainstream drone detection technologies mainly rely on single or limited-dimensional detection equipment, which is insufficient to meet the needs of defense in complex areas. Drone detection technologies have the following limitations:

[0004] The detection dimension is limited, resulting in a high rate of false negatives and false positives.

[0005] Existing technologies mostly employ single radar detection or radio spectrum monitoring, such as capturing drone location information solely through millimeter-wave radar or identifying communication signals solely through radio modules. However, low-altitude, slow-moving, and small drones (e.g., weighing <2kg and flying at an altitude <100m) have extremely small radar cross-sections (RCS) (typically ≤0.01㎡), making them easily obstructed by buildings, trees, etc., leading to missed detections by radar. At the same time, radio signals from civilian devices, including mobile phones and wireless cameras, are easily confused with drone communication signals, resulting in a false detection rate of over 15% for radio spectrum monitoring, making it difficult to accurately identify true drone targets.

[0006] Lack of regionally differentiated detection strategies and unreasonable resource allocation:

[0007] Existing detection systems mostly adopt a uniform deployment model across the entire area, without dividing the defense zone into sub-zones according to the security level. This results in insufficient detection equipment density in core sensitive areas, while equipment is redundant in peripheral non-sensitive areas, leading to wasted resources. For example, an airport's defense system uses the same density of radars around the runway and parking lot, which fails to meet the high-sensitivity detection requirements of the runway area (requiring positioning accuracy ≤0.5m) and results in equipment utilization of less than 30% in the parking lot area.

[0008] Weak multi-source data fusion capability and low data value utilization rate:

[0009] Although some detection systems integrate multiple types of equipment such as radar, optoelectronics, and acoustics, they lack an effective data fusion mechanism: the timestamps of radar data and optoelectronic data are not synchronized, and acoustic data is used directly without filtering environmental noise, which makes it difficult for multi-source data to form a synergy and to extract accurate UAV features (such as flight speed change rate and acoustic spectrum peak), affecting the accuracy of subsequent threat assessment.

[0010] Existing drone threat classification technologies have shortcomings. Threat classification is a prerequisite for achieving precise control, but existing technologies have significant deficiencies in classification logic and model adaptability:

[0011] The grading standard is too simplistic and does not take into account regional safety levels.

[0012] Existing methods often rely solely on simple parameters such as the drone's flight speed and whether it carries a camera to determine threat levels, without considering the security priority of the area where the drone is located. For example, the same civilian aerial photography drone may be considered harmless in an ordinary park area, but should be considered suspicious in a sensitive core area. Current systems do not implement such differentiated classification, which can easily lead to misjudgment of threats in core areas or over-control of peripheral areas.

[0013] In the control and execution phase, existing technologies struggle to achieve tiered control and resource optimization. The specific shortcomings of existing drone control technologies include:

[0014] Control measures are not commensurate with the threat level, resulting in low response efficiency.

[0015] Existing systems often employ a one-size-fits-all approach to control, such as initiating electromagnetic interference on all intruding drones without differentiating actions based on threat levels. For harmless civilian drones, excessive interference can disrupt normal communication; for high-risk attack drones, relying solely on a single interference measure (without combining it with physical interception) is insufficient to effectively prevent intrusion. Furthermore, the response time of existing systems is generally ≥5 seconds, which cannot meet the emergency response requirements of ≤1 second in core areas.

[0016] Weak cross-regional coordination capabilities and inadequate response to multi-target intrusions:

[0017] When multiple drones simultaneously invade different sub-regions, the existing system lacks a unified resource scheduling mechanism. The control equipment in each region operates independently, which can easily lead to insufficient resources in the core area and idle resources in the outer area, resulting in a multi-target capture success rate of less than 70%. At the same time, cross-regional data sharing suffers from high latency.

[0018] Therefore, there is an urgent need to improve the existing drone detection and control process in order to solve the technical problems existing in the above-mentioned technologies. Summary of the Invention

[0019] The purpose of this invention is to provide a method and system for detecting and controlling unmanned aerial vehicles (UAVs) for regional defense, in order to solve the technical problems existing in the prior art.

[0020] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0021] The first aspect provides a method for detecting and controlling drones for regional defense, including the following steps:

[0022] S1: Divide the area to be defended into at least one core defense sub-area and at least two peripheral early warning sub-areas, and set the drone no-fly threshold and detection response priority for each sub-area;

[0023] S2: Construct a multi-dimensional collaborative detection network based on sub-region division: Detection equipment is deployed differently in each sub-region, and the detection equipment in each sub-region collects multi-source data of each target UAV in real time within the corresponding spatial range;

[0024] S3: Based on multi-source data, determine the sub-region where the drone is located and whether it exceeds the no-fly threshold of that sub-region, obtain target data that exceeds the threshold, extract features from the target data, and obtain target feature sets for each sub-region, including flight speed, altitude, communication protocol type, radar cross section and acoustic information;

[0025] S4: Input the extracted feature set into the target dynamic classification model, and the target dynamic classification model outputs the threat level of each target UAV in real time based on the feature set;

[0026] S5: Implement a hierarchical dynamic control strategy based on the detection response priority of each sub-region and the corresponding target drone threat level;

[0027] S6: After implementing the hierarchical dynamic control strategy, continuously collect the status parameters of each target UAV, dynamically evaluate the control effect based on the detection response priority of each sub-region and the preset evaluation indicators, and adaptively adjust the detection equipment parameters and control strategy of each sub-region according to the evaluation results.

[0028] Preferably, the specific process of step S1 is as follows:

[0029] S11: Divide the area to be defended into spatial sub-areas, and determine at least one core defense sub-area and at least two peripheral early warning sub-areas;

[0030] S12: Set no-fly thresholds for drones for each sub-region;

[0031] S13: Set the detection response priority for each sub-region. The core defense sub-region has the highest priority, the inner outer sub-region has the medium priority, and the outer outer sub-region has the basic priority.

[0032] Preferably, the specific process of step S2 is as follows:

[0033] S21: Based on each sub-region and its corresponding detection priority, detection equipment is deployed in each sub-region in a differentiated manner to build a multi-dimensional collaborative detection network, including radio spectrum monitoring modules, millimeter-wave radar arrays, photoelectric tracking equipment and acoustic sensors;

[0034] S22: The radio spectrum monitoring module uses frequency hopping scanning technology to capture radio spectrum monitoring data, including UAV communication frequency band signals, in real time;

[0035] Millimeter-wave radar arrays achieve dynamic tracking of low, slow, and small targets and acquire millimeter-wave radar data through adaptive beamforming algorithms;

[0036] The photoelectric tracking device acquires photoelectric data containing the visual positioning information of the UAV;

[0037] The acoustic sensor uses an acoustic signature database to identify the propeller noise of the drone and obtain acoustic sensor data.

[0038] Preferably, the specific process of step S3 is as follows:

[0039] S31: Perform coordinate calibration by comparing the coordinates of the sub-regional boundaries with the preset coordinates of the GIS geographic information system of the defense area;

[0040] S32: Match the calibrated UAV real-time coordinates with the boundaries of each sub-region:

[0041] S33: For drone data with added sub-region affiliation tags, real-time comparison is performed from three dimensions: altitude, speed, and dwell time. If any of the dimensions exceeds the standard, it is marked as an over-limit target and its original data is included in the over-limit target dataset; data that does not exceed the standard is marked as a compliant target, only the basic record is retained, and it is not included in the subsequent feature extraction process.

[0042] S34: Preprocess the dataset of out-of-limit targets, remove interfering data, and perform spatiotemporal alignment of the data;

[0043] S35: Based on the preprocessed target dataset, extract the corresponding features from each type of data to generate the target feature set for each sub-region.

[0044] Preferably, the feature extraction process for each type of data in step S35 is as follows:

[0045] Instantaneous velocity is calculated from millimeter-wave radar data, and flight velocity features including mean and variance are extracted. RCS values ​​are calculated based on radar echo power, and radar cross section RCS features including mean and fluctuation range are extracted.

[0046] For millimeter-wave radar data and optoelectronic data, obtain radar altitude H1 and optoelectronic altitude H2, and extract flight altitude features such as maximum altitude and duration of continuous exceedance.

[0047] For radio spectrum monitoring data, communication protocol type features are extracted by analyzing signal modulation methods and frame structures and matching protocol libraries.

[0048] For acoustic sensor data, the peak frequency, bandwidth, and period of the noise signal are extracted and matched with the UAV voiceprint database to obtain voiceprint information features.

[0049] Preferably, the target dynamic classification model is based on traditional CNN, with the addition of a sub-region feature attention module and a multi-dimensional feature fusion layer, and the architecture consists of 3 layers:

[0050] Convolutional layer: Three convolutional kernels are used to extract features from the standardized input vector, capture local feature associations, and output a feature map;

[0051] Attention layer: Based on the sub-region affiliation label, feature weights are dynamically assigned. The core area feature set is given 2 times the weight for communication protocol encryption level and voiceprint anomaly, while the outer area is given 1.5 times the weight for speed exceeding the standard and dwell time exceeding the standard, ensuring that key features in high-priority areas are identified first.

[0052] Fully connected layer: The feature maps output by the convolutional layer and attention layer are flattened into a 128-dimensional vector, and then non-linearly transformed through two fully connected layers to finally output a 3-dimensional vector, corresponding to the probability values ​​of the three threat levels.

[0053] Preferably, the specific process of step S4 is as follows:

[0054] S41: Perform a structured transformation on the target feature set, converting it into a numerical matrix format recognizable by the target dynamic classification model:

[0055] S42: The input vector is normalized using the Z-Score normalization algorithm;

[0056] S43: Taking a target feature set in the core area as an example, the convolutional kernel calculates the local feature response through a sliding window, and the model generates the attention weight matrix through the Sigmoid function; during the model training phase, the error between the predicted value and the true label is calculated through the cross-entropy loss function, and the Adam optimizer is used to adjust the parameters of each layer in reverse.

[0057] S44: Threat Level Probability Calculation:

[0058] The 3D vector output by the fully connected layer is converted into probability values ​​for each threat level using the Softmax function, satisfying a probability sum of 1:

[0059] S45: The threat level is determined by combining the maximum probability priority rule with the sub-regional differentiation threshold rule, and the final threat level is output:

[0060] Preferably, the specific process of step S5 is as follows:

[0061] S51: Set the correspondence between sub-region priorities and resource allocation, and set the priority-threat level management strategy decision matrix.

[0062] S52: Perform differentiated actions based on priority-threat level matching:

[0063] Low-intensity control measures will be implemented for harmless civilian drones, medium-intensity control measures will be implemented for suspicious reconnaissance drones, and high-intensity control measures will be implemented for high-risk attack drones.

[0064] Preferably, the specific process of step S6 is as follows:

[0065] S61: Set differentiated evaluation indicators for different control measures:

[0066] Early warning indicators: Departure rate of harmless civilian drones, command reception success rate;

[0067] Interference-related indicators: Image transmission interruption rate and false interference rate of suspicious reconnaissance UAVs;

[0068] Interception metrics: Success rate of capturing high-risk attack drones, interception response time;

[0069] S62: Based on the assessment results, adjust the control strategy and resource allocation in reverse:

[0070] If the success rate of intercepting high-risk attack drones in the core area is less than 99%, increase the number of drones to be intercepted or increase the diameter of the capture net.

[0071] If the false interference rate of suspicious reconnaissance drones in the inner and outer perimeter areas is higher than 0.1%, optimize the electromagnetic interference waveform generation algorithm and increase the identification and filtering of signals from civilian equipment;

[0072] If the departure rate of harmless civilian drones in the outer perimeter area is less than 90%, the early warning method will be upgraded and the early warning range will be expanded.

[0073] S63: Establish an emergency response plan update mechanism:

[0074] The priority-threat level decision matrix is ​​updated monthly based on historical control data.

[0075] If a new type of drone is found to be easily misjudged as harmless in the core area, its RCS feature weight will be increased, while the threshold for judging high-risk attack drones will be lowered.

[0076] If multiple coordinated intrusions occur repeatedly in the outer perimeter area, a new multi-target linkage contingency plan for the outer perimeter area will be added, and a resource allocation path for adjacent areas will be preset to shorten the support response time.

[0077] Secondly, a drone detection and control system for regional defense is used to implement the aforementioned drone detection and control method for regional defense. It includes a regional division and parameter setting module, a multi-dimensional collaborative detection network module, a target data processing and feature extraction module, a target dynamic classification model, a hierarchical dynamic control execution module, and a control effect evaluation and strategy optimization module. Each module achieves data interaction and collaborative work through a 5G private network or fiber optic network.

[0078] The area division and parameter setting module is used to spatially divide the area to be defended and preset the parameters.

[0079] The multi-dimensional collaborative detection network module, based on the output of the region division and parameter setting module, deploys detection devices in different sub-regions to collect multi-source data of the target UAV in real time.

[0080] The target data processing and feature extraction module is used to receive multi-source data transmitted by the multi-dimensional collaborative detection network module and perform data filtering, preprocessing and feature extraction.

[0081] The target dynamic classification model is used to receive the target feature set from the target data processing and feature extraction module to determine the threat level of the UAV.

[0082] The hierarchical dynamic control execution module is used to perform differentiated control operations based on the response priority of the regional division and parameter setting module and the threat level of the target dynamic classification module.

[0083] The control effect evaluation and strategy optimization module is used to continuously collect post-control data from the hierarchical dynamic control execution module to achieve effect evaluation and strategy adjustment.

[0084] The beneficial effects of this invention include:

[0085] 1. The area to be defended is divided into core defense sub-regions and peripheral early warning sub-regions, with differentiated no-fly thresholds and detection response priorities set, avoiding the resource waste caused by the uniform deployment across the entire area in existing technologies. This improves equipment utilization, ensuring high security protection for core sensitive areas while avoiding equipment redundancy in peripheral non-sensitive areas. A three-tiered defense line—core defense, inner early warning, and outer early warning—is constructed through sub-regional division. Relay control is initiated for drones moving across sub-regions (outer early warning data is synchronized to the inner layer, and the inner layer preheats its equipment), solving the problem of weak cross-regional coordination in existing technologies. This forms a three-dimensional defense system with outer early warning, inner interception, and core backup.

[0086] 2. Construct a multi-dimensional collaborative detection network integrating radio spectrum monitoring, millimeter-wave radar, photoelectric tracking, and acoustic sensors. Through coordinate calibration and spatiotemporal alignment, achieve deep fusion of multi-source data to form detection coverage without blind spots or dead zones. Compared to existing single detection technologies, the false negative and false positive rates are reduced.

[0087] 3. The target dynamic classification model assigns a 2x weight to the communication protocol encryption level and voiceprint anomaly of targets in the core area, and a 1.5x weight to the speed exceeding the limit and the dwell time of targets in the peripheral area. This solves the misjudgment problem caused by the existing technology's classification not being combined with the regional security level. The misjudgment rate of threat level in the core area is reduced, avoiding both over-control of harmless core areas being mistakenly identified as high-risk, and preventing high-risk vulnerabilities from being mistakenly identified as harmless.

[0088] 4. By employing a priority-threat level decision matrix, low-intensity early warnings are implemented for harmless civilian drones, medium-intensity jamming is applied to suspicious reconnaissance drones, and high-intensity interception is carried out for high-risk attack drones, thus addressing the problem of the current one-size-fits-all approach to control. The core area control response time is strictly controlled to ≤1 second. Through high-bandwidth data transmission in the core area and high-power jamming in the core area, combined with the coordinated operation of multiple interceptor drones, the slow response time of existing technologies is resolved. For malicious intrusion incidents in the core area, the success rate of control is improved, providing second-level response security for core sensitive areas and effectively preventing high-risk drone attacks.

[0089] 5. Precise indicators are set for different sub-regions and control types to address the lack of standardized assessment methods in existing technologies. Parameters are adjusted in reverse based on assessment results, and emergency plans are updated monthly to overcome the passive state of execution-failure-repair in existing technologies. The system uses historical data backtracking to uncover intrusion patterns, optimizes device deployment and control strategies, and improves overall defense efficiency monthly. It maintains long-term adaptability to complex intrusion scenarios, avoids a decline in defense capabilities due to environmental changes or technological iterations, and forms a closed loop of detection-classification-control-assessment-optimization. Attached Figure Description

[0090] Figure 1 This is a flowchart illustrating the drone detection and control method for regional defense according to the present invention.

[0091] Figure 2 This is a schematic diagram of the defense area division of the present invention.

[0092] Figure 3 This is a schematic diagram illustrating the principle of the target dynamic classification model of the present invention. Detailed Implementation

[0093] The following is in conjunction with the appendix Figures 1-3 The present invention will be further described in detail below:

[0094] Example 1

[0095] See appendix Figure 1 As shown, a method for detecting and controlling drones for area defense includes the following steps:

[0096] S1: Divide the defense area into at least one core defense sub-area and at least two peripheral early warning sub-areas. Set the no-fly threshold and detection response priority for drones in each sub-area. The no-fly threshold for drones includes flight altitude, speed, and dwell time.

[0097] S2: Construct a multi-dimensional collaborative detection network based on sub-region division: Detection equipment is deployed differently in each sub-region, and the detection equipment in each sub-region collects multi-source data of each target UAV in real time within the corresponding spatial range.

[0098] S3: Based on the multi-source data, determine the sub-region where the UAV is located and whether it exceeds the no-fly threshold of the sub-region, obtain target data that exceeds the threshold, extract features from the target data, and obtain target feature sets for each sub-region including flight speed, altitude, communication protocol type, radar cross section and acoustic information.

[0099] S4: Input the extracted feature set into the target dynamic classification model. The target dynamic classification model outputs the threat level of each target UAV in real time based on the feature set. The threat level includes harmless civilian type, suspicious reconnaissance type, and high-risk attack type. The threat level of target UAVs exceeding the threshold in the core defense sub-region is increased by one level by default.

[0100] S5: Implement a tiered dynamic control strategy based on the detection response priority of each sub-region and the corresponding target drone threat level.

[0101] S6: After implementing the hierarchical dynamic control strategy, continuously collect the status parameters of each target UAV, dynamically evaluate the control effect based on the detection response priority of each sub-region and the preset evaluation indicators, and adaptively adjust the detection equipment parameters and control strategy of each sub-region according to the evaluation results.

[0102] In this embodiment, the specific process of step S1 is as follows:

[0103] S11: See also Figure 2 The area to be defended is spatially divided, and at least one core defense sub-area and at least two peripheral early warning sub-areas are identified. The two peripheral early warning sub-areas are the inner peripheral sub-area within a range of 1-3 km outside the core defense sub-area, and the outer peripheral sub-area within a range of 3-10 km outside the core defense sub-area.

[0104] S12: Set no-fly thresholds for drones for each sub-area. The no-fly thresholds for the core defense sub-area are: flight altitude ≤ 0m (complete no-fly), no speed limit (the highest level response is triggered once entered), and dwell time ≤ 0s (dwelling is prohibited).

[0105] The no-fly threshold for the outer warning sub-area is:

[0106] The flight altitude of the inner outer sub-region is ≤50m, the speed is ≤30km / h, and the dwell time is ≤5min; the flight altitude of the outer outer sub-region is ≤120m, the speed is ≤60km / h, and the dwell time is ≤15min.

[0107] S13: Set the detection response priority for each sub-region. The core defense sub-region has the highest priority and a response delay of ≤10s. The inner outer sub-region has a medium priority and a response delay of ≤2s. The outer outer sub-region has a basic priority and a response delay of ≤3s.

[0108] The specific process of step S2 is as follows:

[0109] S21: Based on each sub-region and its corresponding detection priority, detection devices are deployed in each sub-region in a differentiated manner to build a multi-dimensional collaborative detection network;

[0110] In the core defense sub-area, a density of 2-3 radio spectrum monitoring modules per square kilometer, 1-2 sets of millimeter-wave radar arrays per square kilometer, photoelectric tracking equipment with no blind spots, and acoustic sensors are deployed.

[0111] In the outer warning sub-regions, the equipment density is reduced in order of priority, with the equipment density in the inner outer sub-regions being 1 / 2 that of the core region and the equipment density in the outer outer sub-regions being 1 / 3 that of the core region.

[0112] S22: The radio spectrum monitoring module uses frequency hopping scanning technology to capture radio spectrum monitoring data, including UAV communication frequency band signals, in real time;

[0113] The millimeter-wave radar array achieves dynamic tracking of low, slow, and small targets through an adaptive beamforming algorithm, and collects millimeter-wave radar data including the real-time three-dimensional coordinates (longitude X, latitude Y, and altitude H) of the UAV.

[0114] The photoelectric tracking device acquires photoelectric data containing the visual positioning information of the UAV, including the azimuth and distance of the UAV relative to the reference object, to help verify the accuracy of the radar coordinates. When the radar signal is blocked, it automatically switches to photoelectric coordinates as the basis for sub-region matching.

[0115] The acoustic sensor identifies the propeller noise of the drone based on the acoustic signature database to obtain acoustic sensor data. Each detection device achieves synchronous data transmission through a 5G private network, and the data transmission bandwidth and processing priority of high-priority sub-regions are higher than those of low-priority sub-regions.

[0116] Example 2

[0117] Based on Example 1, the specific process of step S3 is as follows:

[0118] S31: By calibrating the coordinates of the sub-region boundaries with the preset sub-region boundary coordinates of the defense area GIS geographic information system, the boundaries include the latitude and longitude range of the polygonal / ring boundary of the core defense sub-region and the outer early warning sub-region, eliminating terrain elevation differences, and the height value can be corrected by the digital elevation model (DEM).

[0119] S32: Match the calibrated UAV real-time coordinates with the boundaries of each sub-region:

[0120] If the coordinates fall within the boundary of the core defense sub-region, such as within the latitude and longitude range of N30°00′-N30°05′ and E120°00′-E120°05′ of the core region, it is determined to be a target in the core region.

[0121] If the coordinates fall within the inner outer sub-region, that is, the 1-3km ring range outside the core area, it is determined to be an inner outer target;

[0122] If the coordinates fall within the outer perimeter sub-region, within a 3-10km ring range outside the core area, it is determined to be an outer perimeter target.

[0123] Add sub-region affiliation tags to each target drone and call up the corresponding no-fly parameters for the sub-region;

[0124] S33: Real-time comparison of drone data with added sub-region affiliation tags, based on three dimensions: altitude, speed, and dwell time.

[0125] Height comparison: The height value H collected in real time by millimeter-wave radar is compared with the height threshold of the sub-region. For example, if H > 0m in the core area and H > 50m in the inner and outer perimeter, it is determined that the height exceeds the standard.

[0126] Speed ​​comparison: The instantaneous speed V of the UAV is calculated by radar Doppler effect, in km / h, and compared with the speed threshold of the sub-region. If V > 30 km / h for the inner and outer regions, the speed is judged to be out of the limit.

[0127] Dwell time comparison: Record the time T1 when the drone first enters the sub-area, and calculate the dwell time ΔT=T in real time. 当前 -T1 is compared with the sub-region stay time threshold. If the outer perimeter ΔT>15min, it is determined that the stay exceeds the standard.

[0128] If any dimension exceeds the limit, the target in the core area will be marked as an over-limit target upon entry, and the original data, including coordinates, speed, height, and timestamp, will be included in the over-limit target dataset; data that does not exceed the limit will be marked as compliant targets, and only basic records will be retained, without proceeding to the subsequent feature extraction stage.

[0129] S34: Preprocess the dataset containing out-of-limit targets to eliminate interference data:

[0130] Radio spectrum monitoring module data: Removes noise signals within the frequency band, such as civilian broadcast signals and interference signals from other electronic devices. By filtering through signal power thresholds, only drone communication signals with power ≥ -110dBm are retained.

[0131] Acoustic sensor data: Wavelet transform denoising algorithm is used to filter environmental noise, such as wind noise and vehicle noise, while retaining the characteristic frequency band of drone propeller noise in the range of 200-2000Hz.

[0132] Radar and photoelectric data: Outliers, such as instantaneous height jumps and speed anomalies caused by signal reflection, are removed. The data is smoothed using a moving average method with a window size of 5 sampling points to ensure data continuity.

[0133] Perform spatiotemporal alignment of data:

[0134] Because different detection devices have different sampling frequencies (e.g., radar sampling interval is 10ms, photoelectric sampling interval is 30ms), a timestamp synchronization algorithm is used, based on the unified clock of the 5G private network, to align multi-source data to the same time dimension, such as generating one fused data record every 10ms. Simultaneously, the spatial coordinates of each device are uniformly converted to the local coordinate system of the defense area, such as a Cartesian coordinate system with the center point of the core area as the origin, ensuring data spatial consistency.

[0135] S35: Based on the preprocessed target dataset, extract the corresponding features from each type of data to generate the target feature set for each sub-region.

[0136] The feature extraction results for each out-of-limit target are used to generate a structured feature set according to the sub-region type, as shown in the following example format:

[0137] Core defense sub-region feature set: {Target ID: 001, Sub-region label: Core region, Velocity feature: [V] 均值 =25km / h, ΔV / Δt=3km / h・s], Altitude characteristics: [H max =10m, duration of exceeding limit =12s], communication protocol: dedicated encryption protocol, RCS: 0.01㎡, voiceprint anomaly score: 85 points (out of 100)};

[0138] The feature set includes sub-region identifiers, such as core area feature set-20250925-001, which are transmitted in real time to the target dynamic classification model as core input data for threat level determination. At the same time, the feature sets of each sub-region are stored in the database in time series for subsequent control effect evaluation and intrusion pattern analysis.

[0139] In this embodiment, the feature extraction process for each type of data in step S35 is as follows:

[0140] Instantaneous velocity is calculated from millimeter-wave radar data, and flight velocity features including mean and variance are extracted. The RCS value is calculated based on radar echo power, in square meters, and radar cross section RCS features including mean and fluctuation range are extracted.

[0141] For millimeter-wave radar data and photoelectric data, radar altitude H1 and photoelectric altitude H2 are obtained, with a weight ratio of 6:4. Flight altitude features with maximum altitude and duration of continuous exceedance are extracted.

[0142] For radio spectrum monitoring data, communication protocol type characteristics are extracted by analyzing signal modulation methods and frame structures and matching with protocol libraries including WiFi, Bluetooth, ZigBee, and proprietary protocols.

[0143] For acoustic sensor data, the peak frequency, bandwidth, and period of the noise signal are extracted and matched with the UAV voiceprint database to obtain voiceprint information features.

[0144] Example 3

[0145] Based on Example 1 or Example 2, see Figure 3 The target dynamic classification model is based on traditional CNN, with the addition of a sub-region feature attention module and a multi-dimensional feature fusion layer. The architecture consists of three layers:

[0146] Convolutional Layer (Conv Layer): Three convolutional kernels with sizes of 3×1, 5×1, and 7×13 are used to extract features from the standardized input vector, capture local feature associations, including the synergistic relationship between the rate of change of velocity and the rate of high approximation, and output a 64-dimensional feature map.

[0147] Attention Layer: Based on the sub-region affiliation label, feature weights are dynamically assigned. The core region feature set is given twice the weight of the communication protocol encryption level and the voiceprint anomaly degree, with a weight coefficient of 2.0. The outer region is given 1.5 times the weight of the speed exceeding the standard and the dwell time exceeding the standard, ensuring that key features in high-priority regions are identified first.

[0148] Fully Connected Layer (FC Layer): The feature maps output by the convolutional layer and attention layer are flattened into a 128-dimensional vector. The vector is then transformed non-linearly by two fully connected layers with 64 and 32 hidden units respectively, and finally outputs a 3-dimensional vector, corresponding to the probability values ​​of the three threat levels.

[0149] In this embodiment, the specific process of step S4 is as follows:

[0150] S41: Perform a structured transformation on the target feature set, converting it into a numerical matrix format recognizable by the target dynamic classification model:

[0151] For example, the core region feature set in the target feature set output in step S3 is: {Target number: 001, velocity feature: [V]} 均值 =25km / h, ΔV / Δt=3km / h・s], Altitude characteristics: [H max =10m, duration of exceeding limit =12s], communication protocol: dedicated encryption protocol, RCS: 0.01㎡, voiceprint anomaly score: 85 points} Convert to a numerical matrix format recognizable by the model:

[0152] Numerical features are directly quantified: the continuous values ​​of mean speed (25), speed change rate (3), maximum height (10), duration of exceeding the standard (12), RCS value (0.01), and voiceprint anomaly degree (85) are directly retained as the original data;

[0153] Categorical feature encoding conversion: Communication protocol types are converted into binary vectors using one-hot encoding, such as encryption protocol → [1,0,0,0], WiFi → [0,1,0,0], to ensure that the model can perform numerical calculations.

[0154] Generate input vector: Arrange all features in a fixed order of speed feature → height feature → communication protocol feature → RCS feature → voiceprint feature to form an input vector with dimension 1×N, where N is the total number of features, such as N=8 for the core area feature set and N=6 for the outer area feature set.

[0155] S42: To eliminate the impact of differences in the units of different feature dimensions on model inference, such as speed in km / h and RCS in m², the Z-Score normalization algorithm is used to normalize the input vector:

[0156] Calculation formula: x norm =( x−μ ) / σ ,in x These are the original eigenvalues. μ This is the mean of the feature in the training dataset. σ The standard deviation is denoted as .

[0157] Standardization range: Map all feature values ​​to the interval [-1, 1]. For example, if the original value of voiceprint anomaly is 85, μ=50. σ =20, after standardization it becomes (85−50) / 20=1.75, and then the truncation process is used to limit it to within 1 to ensure the stability of the model input data.

[0158] Sub-region differentiation parameters: Standardized parameters are calculated separately for the feature sets of the core region and the outer region, such as the RCS training set of the core region. μ =0.008㎡, outer area μ =0.015㎡, to avoid errors caused by differences in cross-regional characteristic distribution.

[0159] S43: Local Feature Extraction (Convolutional Layer): Taking a target feature set in the core area as an example, the convolutional kernel calculates the local feature response through a sliding window with a stride of 1. For example, a 3×1 convolutional kernel calculates the associated features of "average velocity - rate of change of velocity - maximum height". The output feature value reflects the degree of coordination anomaly among the three. For example, when the velocity increases sharply and the height approaches the core area, the feature value increases significantly.

[0160] Attention weight adjustment: The model generates an attention weight matrix using the Sigmoid function. The feature weights for the core target "communication protocol encryption level" are calculated as w= σ (a×x+b), where a and b are feature weight coefficients, a=0.8, b=0.2, and x is the sub-region identifier vector. If it is a core area target, x=1, then w= σ (0.8×1+0.2)=0.73, then multiplied by the base weight of 2.0, the final weight is 1.46.

[0161] Gradient backpropagation optimization: During the model training phase, the error between the predicted value and the true label (manually labeled drone threat level) is calculated using the cross-entropy loss function. The Adam optimizer (learning rate 0.001) is used to backpropagate and adjust the parameters of each layer to ensure that the prediction accuracy of the core area threat level is ≥99.5% and the outer area is ≥99.2%.

[0162] S44: Generating threat level probability values ​​using the Softmax function:

[0163] The 3D vector output by the fully connected layer (e.g., [0.12, 0.35, 0.53]) is converted into probability values ​​for each threat level using the Softmax function, satisfying a probability sum of 1:

[0164] Calculation formula: ,in zi The first output of the fully connected layer i A vector value, where j is the category number of the threat level, j=1, 2, 3, zj The original fraction;

[0165] For example: when the output of the fully connected layer is [0.2, 0.5, 0.8], the probabilities of each level are:

[0166] Harmless civilian type P1= e 0.2 / ( e 0.2 + e 0.5 + e 0.8 ) ≈1.22 / (1.22+1.65+2.23)≈23%;

[0167] Suspicious reconnaissance type P2 = 1.65 / 5.1 ≈ 32%;

[0168] High-risk attack type P3 = 2.23 / 5.1 ≈ 44%.

[0169] S45: The threat level is determined by combining the maximum probability priority rule with the sub-regional differentiation threshold rule (see Table 1), and the final threat level is output:

[0170] Table 1

[0171]

[0172] Set up special case handling strategies:

[0173] If a target in the core area uses a dedicated encrypted protocol and has an abnormal voiceprint score of ≥80, or ≥1.0 after standardization, it will be forcibly upgraded to a suspicious reconnaissance target even if P3 <0.4. If a target in the outer area exceeds the speed limit by ≥50%, such as an inner-outer speed of ≥45km / h, the P2 weight will be increased by 0.1 to avoid misjudgment.

[0174] Setting up the optimization process for the target dynamic classification model:

[0175] Novel UAV Feature Adaptive Learning: When the model encounters an extracted set of unknown features, such as the unique communication protocol and RCS features of the novel UAV, a real-time learning process is initiated.

[0176] Feature anomaly detection: By calculating the Euclidean distance between the input feature and the training set features, if the RCS value deviates from the standard deviation of the training set by ≥3σ, it is determined to be an unknown feature, and the target is marked as a sample to be learned.

[0177] Edge computing fine-tuning: A small number of labeled samples are generated on local edge computing nodes, and combined with manual annotation, such as new suspicious drones, the parameters of the model's convolutional layers are frozen using transfer learning techniques, and only the parameters of the fully connected layers and attention layers are fine-tuned.

[0178] Update cycle: The core area is triggered when there are ≥5 learning samples, with a cycle of ≤30 minutes. The outer area is triggered when there are ≥10 learning samples, with a cycle of ≤1 hour. After the update, the model is automatically synchronized to the detection nodes of each sub-region.

[0179] Historical data feedback optimization:

[0180] Every day at midnight, a retrospective analysis is performed on the feature set, threat level, and actual control results data from the previous 24 hours:

[0181] Error statistics: Calculate the false alarm rate of threat level in each sub-region, such as the proportion of high-risk attack type being mistaken for suspicious reconnaissance type in the core region. If the false alarm rate is ≥0.5%, adjust the attention weight of the corresponding sub-region, such as increasing the RCS feature weight in the core region.

[0182] Threshold correction: If a certain type of drone (such as a small foldable reconnaissance drone) is frequently misjudged in the core area, the Softmax judgment threshold of the target feature is recalculated. For example, the P3 judgment threshold is reduced from 0.4 to 0.35 to ensure that the model is adapted to the actual application scenario.

[0183] The specific process of step S5 is as follows:

[0184] S51: Set the correspondence between sub-region priorities and resource configurations, see Table 2 below:

[0185] Table 2

[0186]

[0187] Set the priority-threat level management strategy decision matrix, see Table 3 below:

[0188] Table 3

[0189]

[0190] S52: Perform differentiated actions based on priority-threat level matching:

[0191] Management of Harmless Civilian Drones (Low-Intensity Management):

[0192] For civilian drones that do not carry threatening equipment and only slightly exceed the limits, control measures based on early warning and guidance will be implemented. The specific process is as follows:

[0193] The execution process in the core area is as follows:

[0194] After receiving the harmless civilian-type results output by the model, the geofencing system is activated within 1 second to generate an electronic fence for the core area boundary.

[0195] The ADS-B (Automatic Dependent Surveillance-Broadcast) system sends bilingual (Chinese / English) instructions to the drone, including the coordinates of the no-fly zone and the consequences of violations, at a frequency of once per second to ensure that the operator receives the instructions.

[0196] The photoelectric tracking device continuously locks onto the drone and monitors in real time whether it leaves as instructed. If it does not leave within 30 seconds, it is automatically upgraded to a suspicious reconnaissance-type control.

[0197] The execution process in the outer area is as follows:

[0198] Inner outer perimeter area: Within 2 seconds, a warning command is sent via ADS-B, and the acoustic sensor (sampling rate increased to 48kHz) tracks the drone noise to determine its flight direction;

[0199] Outer perimeter zone: Within 3 seconds, a warning SMS is sent to the operator's mobile phone number through the drone management platform, and the drone's position is recorded (updated every 5 seconds). If the drone moves towards the core zone, an inner zone warning is triggered.

[0200] Management of Suspicious Reconnaissance Drones (Medium-Intensity Management):

[0201] For drones with reconnaissance characteristics (such as carrying high-definition cameras and encrypted communication protocols), implement a combination of jamming, locking, and recording control. Specific steps include:

[0202] Core area execution procedure: Within 1 second, the high-power electromagnetic interference module is activated. Based on the UAV communication frequency band provided by the radio spectrum monitoring module, a directional interference signal with a power of 8-10W is generated. The interference direction is calibrated in real time by millimeter-wave radar, with a deviation of ≤1°. The millimeter-wave radar and optoelectronic equipment lock in dual mode, recording the UAV's flight trajectory, including latitude, longitude, altitude, and speed, storing one data point every 100ms. Simultaneously, ground control personnel (equipped with portable jamming equipment) are notified to proceed to the area where the UAV may be forced to land. If the UAV continues to remain after interference, an anti-UAV swarm is activated within 5 seconds.

[0203] Outer Zone Execution Procedure: Inner Outer Zone: Within 2 seconds, initiate medium-power jamming (3-5W), focusing on jamming the drone's image transmission frequency band (e.g., 5.8GHz), while simultaneously analyzing the flight trajectory for any evasive maneuvering characteristics (e.g., frequent changes of direction). Outer Outer Zone: Within 3 seconds, initiate low-power jamming (0.1-2W), only jamming the remote control frequency band (without affecting surrounding civilian equipment), and upload the trajectory data to the regional joint defense platform to predict whether the drone will move towards the core area.

[0204] High-risk offensive drone management (high-intensity management): For drones carrying offensive equipment (such as explosives, laser emitters) or possessing high-speed penetration capabilities, implement comprehensive management including interception, coordinated action, and emergency response. Specific steps:

[0205] Core Area Execution Procedure: Within 1 second, trigger the anti-drone swarm takeoff command. 3-5 interceptor drones take off from preset takeoff and landing points (5 around the core area, spaced ≤1km apart), calibrating the target position using lidar (detection range ≥1km). The interceptor drones employ a triangular encirclement tactic, forming an interception circle with a diameter of 50m. When the distance to the target is ≤50m, a foldable capture net is deployed, with a capture success rate ≥99%. Simultaneously, the primary ground-based air defense system (such as short-range air defense missiles) enters standby status. If capture fails, the air defense missile lock-on procedure is initiated within 10 seconds, and personnel in the core area are evacuated.

[0206] Outer Zone Execution Procedure: Inner Outer Zone: Within 2 seconds, dispatch 1-3 interceptor drones, employing tail-following interception tactics. Electromagnetic interference modules (3-5W) continuously interfere with the target's navigation signals (GPS / BeiDou), forcing the target to slow down. Outer Outer Zone: Within 3 seconds, call upon backup interceptor drones and simultaneously send support requests to neighboring inner zones. Share target data through blockchain technology to form a cross-regional interception echelon. If the target breaches the inner defense line, the core zone interception plan is automatically triggered.

[0207] Cross-regional collaborative action phase: Addressing multi-target, cross-sub-regional intrusion scenarios:

[0208] Multi-target simultaneous intrusion coordination mechanism:

[0209] When multiple sub-regions simultaneously detect high-risk attack drones, such as one in the core area and two in the inner and outer perimeter areas, a priority-based, resource allocation and coordination process is initiated:

[0210] The core area prioritizes the use of 3 interception drones, while the inner outer area calls upon the remaining 2 drones. At the same time, 1 backup drone is allocated from the outer outer area to support the inner area.

[0211] Electromagnetic interference modules are uniformly scheduled by the regional control center to avoid interference conflicts in the same frequency band, such as interference in the core area at 2.4GHz and interference in the inner layer at 5.8GHz.

[0212] Radar resources are allocated using time slicing technology, with the core area radar scanning once every 5ms and the inner layer radar scanning once every 10ms, ensuring the continuity of monitoring in high-priority areas.

[0213] Cross-sub-region mobile target tracking mechanism: When a drone moves from the outer perimeter area to the core area (cross-sub-region intrusion), a relay control process is executed:

[0214] When the target leaves, the outer perimeter area synchronizes its flight trajectory and characteristic parameters (RCS, voiceprint) to the inner perimeter area via a 5G private network, with a synchronization time of ≤1s;

[0215] The electromagnetic interference module in the inner outer area is preheated (power adjusted to 5W) and the interception drone is put on standby. When the target enters the inner area, medium-level control is directly executed without re-identification.

[0216] If the target breaches the inner area, the core area will automatically upgrade its control intensity (e.g., increasing the number of intercepted drones to 5), forming a three-tiered defense line of outer early warning, inner interception, and core defense.

[0217] The specific process of step S6 is as follows:

[0218] S61: Set differentiated evaluation indicators for different control measures to ensure that the control effects can be quantified.

[0219] Early warning indicators: Departure rate of harmless civilian drones: core area ≥98%, inner layer ≥95%, outer layer ≥90%, command reception success rate ≥99%;

[0220] Interference-related indicators: Image transmission interruption rate of suspicious reconnaissance UAVs: core area ≥99%, inner layer ≥97%, outer layer ≥95%, false interference rate ≤0.1%;

[0221] Interception metrics: Success rate of capturing high-risk attack drones: core area ≥99%, inner layer ≥98%, outer layer ≥97%, interception response time ≤1s / 2s / 3s.

[0222] S62: Based on the assessment results, adjust the control strategy and resource allocation in reverse, with specific optimization directions as follows:

[0223] If the success rate of intercepting high-risk attack drones in the core area is less than 99%, increase the number of intercepted drones from 3-5 to 5-7, or increase the diameter of the capture net from 4m to 5m.

[0224] If the false interference rate of suspicious reconnaissance drones in the inner and outer perimeter areas is higher than 0.1%, optimize the electromagnetic interference waveform generation algorithm and increase the identification and filtering of signals from civilian equipment, such as excluding mobile phone communication frequency bands.

[0225] If the departure rate of harmless civilian drones in the outer perimeter area is less than 90%, the early warning method will be upgraded from SMS warning to APP pop-up and telephone reminder. At the same time, the warning range will be expanded from 1km outside the boundary of the outer perimeter area to 2km.

[0226] S63: Establish an emergency response plan update mechanism:

[0227] Monthly updates to the priority-threat level decision matrix based on historical control data, such as new drone intrusion cases and control failures:

[0228] If a new type of drone (such as a micro bionic drone) is found to be easily misjudged as harmless in the core area, its RCS feature weight will be increased from 1.0 to 1.5, while the threshold for judging high-risk attacks will be lowered from P3≥0.4 to P3≥0.35.

[0229] If multiple coordinated intrusions occur repeatedly in the outer perimeter area, a new multi-target linkage contingency plan for the outer perimeter area will be added, with preset resource allocation paths for adjacent areas to shorten the support response time from 3 seconds to 2 seconds.

[0230] The UAV detection and control system for regional defense is used to implement the aforementioned UAV detection and control method for regional defense. It includes a regional division and parameter setting module, a multi-dimensional collaborative detection network module, a target data processing and feature extraction module, a target dynamic classification model, a hierarchical dynamic control execution module, and a control effect evaluation and strategy optimization module. Each module achieves data interaction and collaborative work through a 5G private network or fiber optic network.

[0231] The region division and parameter setting module is used to spatially divide the area to be defended and preset parameters. The multi-dimensional collaborative detection network module, based on the output of the region division and parameter setting module, deploys detection devices in different sub-regions to collect multi-source data of the target UAV in real time. The target data processing and feature extraction module receives multi-source data transmitted from the multi-dimensional collaborative detection network module and performs data filtering, preprocessing, and feature extraction. The target dynamic classification model receives the target feature set from the target data processing and feature extraction module to determine the UAV threat level. The hierarchical dynamic control execution module executes differentiated control operations based on the response priority of the region division and parameter setting module and the threat level of the target dynamic classification module. The control effect evaluation and strategy optimization module continuously collects post-control data from the hierarchical dynamic control execution module to evaluate the effect and adjust the strategy.

Claims

1. A method for detecting and managing unmanned aerial vehicles for area defense, characterized in that, The method comprises the following steps: S1: dividing the defense area to be defended into at least one core defense sub-region and at least two peripheral early warning sub-regions, setting the unmanned aerial vehicle flight prohibition threshold and the detection response priority of each sub-region; S2: constructing a multi-dimensional cooperative detection network based on the sub-region division: differentiating the detection equipment in each sub-region, and each sub-region's detection equipment collecting multi-source data of each target unmanned aerial vehicle in the corresponding space range in real time; S3: judging the sub-region where the unmanned aerial vehicle is located and whether it exceeds the flight prohibition threshold of the sub-region based on the multi-source data, obtaining the target data exceeding the threshold, extracting the features of the target data, and obtaining the target feature set of each sub-region including the flight speed, height, communication protocol type, radar reflection cross section and voiceprint information; S4: inputting the extracted feature set into a target dynamic classification model, and the target dynamic classification model outputting the threat level of each target unmanned aerial vehicle in real time based on the feature set; S5: executing a hierarchical dynamic control strategy according to the detection response priority of each sub-region and the threat level of the corresponding target unmanned aerial vehicle; S6: continuously collecting the state parameters of each target unmanned aerial vehicle after executing the hierarchical dynamic control strategy, dynamically evaluating the control effect based on the detection response priority of each sub-region and the preset evaluation index, and adaptively adjusting the detection equipment parameters of each sub-region and the control strategy according to the evaluation result; The specific process of step S5 is as follows: S51: setting the correspondence between the sub-region priority and the resource configuration, and setting the priority-threat level control strategy decision matrix: S52: executing differentiated operations according to the priority-threat level matching: For harmless civilian unmanned aerial vehicles, low-intensity control is executed; for suspicious reconnaissance unmanned aerial vehicles, medium-intensity control is executed; for high-risk attack unmanned aerial vehicles, high-intensity control is executed.

2. The UAV detection and management method for area defense according to claim 1, wherein, The specific process of step S1 is as follows: S11: spatially dividing the defense area to be defended to determine at least one core defense sub-region and at least two peripheral early warning sub-regions; S12: setting the unmanned aerial vehicle flight prohibition threshold for each sub-region; S13: setting the detection response priority of each sub-region, with the core defense sub-region having the highest priority, the inner peripheral sub-region having a medium priority, and the outer peripheral sub-region having a basic priority.

3. The UAV detection and management method for area defense according to claim 2, wherein, The specific process of step S2 is as follows: S21: based on each sub-region and its corresponding detection priority, differentiating the detection equipment in each sub-region to construct a multi-dimensional cooperative detection network, including a radio spectrum monitoring module, a millimeter wave radar array, an optoelectronic tracking device and an acoustic sensor; S22: the radio spectrum monitoring module uses frequency hopping scanning technology to capture real-time radio spectrum monitoring data including unmanned aerial vehicle communication band signals; The millimeter wave radar array realizes dynamic tracking of low, slow and small targets through an adaptive beam forming algorithm, and collects millimeter wave radar data; The optoelectronic tracking device obtains optoelectronic data containing unmanned aerial vehicle visual positioning information; The acoustic sensor identifies the propeller noise of the unmanned aerial vehicle based on a voiceprint feature library to obtain acoustic sensor data.

4. The UAV detection and management method for area defense according to claim 3, wherein, The specific process of step S3 is as follows: S31: coordinate calibration is performed with the sub-region boundary coordinates preset in the defense area GIS geographic information system; S32: Match the calibrated real-time coordinates of the unmanned aerial vehicle with the boundaries of each sub-region: S33: Real-time comparison of the unmanned aerial vehicle data with the added sub-region attribution labels from the dimensions of height, speed, and duration, and if any dimension exceeds the standard, it is marked as an out-of-limit target, and its original data is included in the out-of-limit target dataset; data that does not exceed the standard is marked as a compliant target, only the basic record is retained, and it does not enter the subsequent feature extraction link; S34: Preprocess the out-of-limit target dataset to eliminate interference data and perform data space alignment; S35: Based on the preprocessed out-of-limit target dataset, extract corresponding features for each type of data to generate target feature sets for each sub-region.

5. The UAV detection and management method for area defense according to claim 4, wherein, The process of extracting corresponding features for each type of data in step S35 is as follows: For millimeter wave radar data, calculate the instantaneous speed, extract the flight speed features including the mean and variance of the speed, and calculate the RCS value based on the radar echo power, extract the radar cross section RCS features including the mean and fluctuation range of the RCS; For millimeter wave radar data and photoelectric data, obtain the radar height H1 and the photoelectric height H2, and extract the flight height features including the maximum height and the duration of exceeding the standard; For radio spectrum monitoring data, extract the communication protocol type features by analyzing the signal modulation method, frame structure, and matching the protocol library; For acoustic sensor data, extract the frequency spectrum peak value, bandwidth, and period of the noise signal, match with the unmanned aerial vehicle voiceprint library, and obtain the voiceprint information features.

6. The UAV detection and management method for area defense according to claim 1, wherein, The target dynamic classification model is based on traditional CNN, with the addition of sub-region feature attention module and multi-dimensional feature fusion layer, and the architecture is divided into 3 layers: Convolution layer: 3 convolution kernels are used to extract features from the standardized input vector, capture local feature associations, and output feature maps; Attention layer: dynamically assign feature weights based on sub-region attribution labels, core area feature set gives 2 times weight to communication protocol encryption level and voiceprint abnormality, and peripheral area gives 1.5 times weight to speed exceeding amplitude and duration exceeding, to ensure that high priority area key features are identified first; Fully connected layer: flatten the feature maps output by the convolution layer and attention layer into 128-dimensional vectors, perform non-linear transformation through 2 fully connected layers, and finally output a 3-dimensional vector corresponding to the probability values of 3 threat levels.

7. The UAV detection and management method for area defense according to claim 1, wherein, The specific process of step S4 is as follows: S41: Structure conversion of target feature set to numerical matrix format recognizable by target dynamic classification model: S42: Normalize the input vector using Z-Score standardization algorithm; S43: Based on a target feature set in the core area, the convolution kernel calculates the local feature response through the sliding window, and the model generates the attention weight matrix through the Sigmoid function; In the model training stage, the error between the predicted value and the true label is calculated by the cross-entropy loss function, and the Adam optimizer is used to adjust the parameters of each layer in reverse; S44: Threat level probability calculation: The 3-dimensional vector output by the fully connected layer is converted to probability values of each threat level by the Softmax function, and the probability sum is 1: S45: The threat level is determined according to the probability maximum priority and the sub-region differentiation threshold rule, and the final threat level is output.

8. The UAV detection and management method for area defense according to claim 1, wherein, The specific process of step S6 is as follows: S61: Differentiated evaluation indexes are set for different control measures: Early warning indexes: harmless civilian UAV departure rate, command reception success rate; Interference indexes: suspicious reconnaissance UAV image transmission interruption rate, false interference rate; Interception indexes: high-risk attack UAV capture success rate, interception response time; S62: According to the evaluation results, the control strategy and resource allocation are adjusted reversely: If the core area high-risk attack UAV interception success rate is lower than 99%, the number of interception UAVs is increased or the capture net deployment diameter is increased; If the inner peripheral area suspicious reconnaissance UAV false interference rate is higher than 0.1%, the electromagnetic interference waveform generation algorithm is optimized, and the identification and filtering of civilian equipment signals are increased; If the outer peripheral area harmless civilian UAV departure rate is lower than 90%, the early warning method is upgraded, and the early warning range is expanded; S63: An emergency plan updating mechanism is set: Based on historical control data, the priority-threat level decision matrix is updated every month: If it is found that a certain type of UAV is easily mistaken for harmless type in the core area, the RCS feature weight is increased, and the high-risk attack judgment threshold is lowered; If multiple targets in the outer peripheral area invade cooperatively for many times, a multi-target linkage plan for the outer peripheral area is added, the resource allocation path in the adjacent area is preset, and the support response time is shortened.

9. An unmanned aerial vehicle detection and control system for area defense, for implementing the unmanned aerial vehicle detection and control method for area defense according to any one of claims 1-8, characterized in that, The system includes a region division and parameter setting module, a multi-dimensional cooperative detection network module, a target data processing and feature extraction module, a target dynamic classification model, a hierarchical dynamic control execution module, and a control effect evaluation and strategy optimization module; each module realizes data interaction and cooperative work through a 5G private network or a fiber network; The region division and parameter setting module is used for spatial division and parameter presetting of the region to be defended; The multi-dimensional cooperative detection network module is based on the output of the region division and parameter setting module, and detection devices are arranged in different sub-regions for real-time collection of multi-source data of target UAVs; The target data processing and feature extraction module is used for receiving multi-source data transmitted by the multi-dimensional cooperative detection network module, and performing data filtering, preprocessing and feature extraction; The target dynamic classification model is used for receiving target feature sets from the target data processing and feature extraction module, and realizing UAV threat level determination; The hierarchical dynamic control execution module is used for executing differentiated control operations according to the response priority of the region division and parameter setting module and the threat level of the target dynamic classification module; The control effect evaluation and strategy optimization module is used for continuously collecting post-control data from the hierarchical dynamic control execution module, and realizing effect evaluation and strategy adjustment.

Images