Unmanned aerial vehicle intrusion detection and security level evaluation method based on visible light vision

Abstract

The invention relates to the technical field of computer vision, and particularly discloses an unmanned aerial vehicle intrusion detection and security level evaluation method based on visible light vision, which comprises the following steps: synchronously acquiring multi-modal visual data of an unmanned aerial vehicle by using a multi-camera array, and carrying out preprocessing and feature extraction on the acquired visual data to obtain key feature information of the unmanned aerial vehicle; constructing a dynamic threat evaluation model based on the key feature information of the unmanned aerial vehicle, and quantifying the threat level of the unmanned aerial vehicle in real time by adopting a weighted Euclidean distance; the unmanned aerial vehicle detection precision of the dynamic threat evaluation model is optimized through a multi-modal feature fusion technology; according to the method, the multi-mode visual data of the unmanned aerial vehicle are synchronously acquired through the multi-camera array, and the threat level is quantified by combining the dynamic threat evaluation model and the weighted Euclidean distance, so that the anti-interference capability of unmanned aerial vehicle detection is remarkably improved; and an optimized visual data preprocessing and feature extraction technology is adopted, so that the detection precision is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision, and particularly relates to a method for detecting drone intrusion and evaluating security level based on visible light vision. Background Art

[0002] With the rapid development and wide application of drone technology, in airspace sensitive areas such as airports and military restricted areas, the security threat problem of drones intruding into sensitive areas has become increasingly prominent. Existing drone intrusion detection technologies mainly include methods such as radio signal monitoring, radar detection, and infrared thermal imaging. However, these technologies have the following limitations: Radio monitoring is easily troubled by electromagnetic interference, and the false alarm rate is as high as 42% in signal-intensive areas such as airport towers; Although radar detection can detect targets at a long distance, it has limited recognition ability for low-altitude micro drones (RCS < 0.5m 2 ), and the equipment cost is high; Infrared thermal imaging depends on temperature difference characteristics, and the detection distance for small targets drops sharply by 70% in low-temperature or foggy environments.

[0003] Visible light vision technology has the advantages of high intuitiveness, low cost, and wide applicability. However, in complex scenarios, it needs to break through two major bottlenecks: one is the background interference problem, such as airport runway stripes and military camouflage, which are likely to cause false detections; the other is the problem of low-altitude target recognition. The characteristics of drones and birds are highly similar. Existing solutions often have difficulty in balancing detection accuracy and anti-interference ability, resulting in limited actual application effects. For example, traditional visible light systems are difficult to adapt to the day-night light changes, and the detection rate drops by more than 50% during sunrise and sunset, and it is easy to miss detections of low-altitude targets in complex terrains.

[0004] Therefore, it is necessary to propose a method for detecting drone intrusion and evaluating security level based on visible light vision to solve the problem of weak anti-interference ability of drone detection in complex environments in the prior art.

[0005] The above information disclosed in this background art is only used to increase the understanding of the background art of the present invention. Therefore, it may include prior art that is not known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for detecting drone intrusion and evaluating security level based on visible light vision to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] A method for detecting drone intrusion and evaluating security level based on visible light vision, comprising:

[0009] Synchronously collect multi-modal visual data of the UAV using a multi-camera array, preprocess and extract features from the collected visual data to obtain key feature information of the UAV;

[0010] Build a dynamic threat assessment model based on the key feature information of the UAV, and use the weighted Euclidean distance to quantify the threat level of the UAV in real time;

[0011] Optimize the UAV detection accuracy of the dynamic threat assessment model through multi-modal feature fusion technology;

[0012] Use the optimized dynamic threat assessment model to evaluate the safety level of the current UAV and generate a safety level assessment result;

[0013] According to the safety level assessment result, combine meteorological data and digital elevation model to update the sensitivity information of the monitoring area in real time.

[0014] Preferably, the multi-camera array adopts a distributed architecture and maintains a stable proportion of the target in the picture by dynamically adjusting the camera focal length; the multi-modal visual data includes the position information, speed information, trajectory deviation information and environmental parameter information of the UAV.

[0015] Preferably, the preprocessing and feature extraction of the collected visual data include:

[0016] Use a spatio-temporal domain joint filtering algorithm to remove noise from the visual data. The formula of the spatio-temporal domain joint filtering algorithm is as follows:

[0017]

[0018] In the formula, E′ is the denoised image, K is the normalization constant, E a,b and G a,b are the image pixel value and the filter kernel value respectively, and S a,b is the spatio-temporal domain weighting coefficient;

[0019] Use an adaptive enhancement algorithm based on the Retinex theory to enhance the area where the target is located;

[0020] Use an improved YOLOv5 network to extract the feature information of the UAV, introduce an attention mechanism to optimize the sensitivity of the YOLOv5 network to details, and obtain the key feature information of the UAV:

[0021]

[0022] In the formula, A(x,y) represents the attention weight, W is the weight matrix, and F(x,y) is the image feature.

[0023] Preferably, the preprocessing and feature extraction of the collected visual data further include:

[0024] Use the hybrid difference algorithm to solve the problems of sudden illumination change and motion speed sensitivity;

[0025] The hybrid difference algorithm combines the background difference method and the three-frame difference method;

[0026] Among them, calculate the difference between the current frame and the background model through the background difference method to generate a background mask;

[0027] Use the three-frame difference method to calculate the difference between the current frame and the previous frame to generate an inter-frame mask;

[0028] Fuse the background mask and the inter-frame mask to obtain a moving target mask;

[0029] Based on the moving target mask, update the background model using the exponential weighted average method;

[0030] Introduce the histogram contrast algorithm to extract the significant region and narrow the range of candidate targets;

[0031] Based on the narrowed range of candidate targets, use the improved YOLOv5 network to extract the key feature information of the UAV.

[0032] Preferably, build a dynamic threat assessment model based on the key feature information of the UAV, and use the weighted Euclidean distance to quantify the threat level of the UAV in real time, including:

[0033] Build the dynamic threat assessment model using a multi-dimensional parameter distance function; the multi-dimensional parameter distance function uses the weighted Euclidean distance formula to quantify the threat level of the UAV in real time based on the position, speed, trajectory deviation degree and environmental parameters of the UAV;

[0034]

[0035] where w r is the weight factor of the rth parameter, h r is the parameter value collected in real time, h r() is the preset safety threshold, and n is the total number of parameters;

[0036] Refer to the spatial coordinates and weight coefficients of the no-fly zone and restricted flight zone in the monitoring area when building the model.

[0037] Preferably, the safety level assessment result is divided into three levels: low, medium, and high; after generating the safety level assessment result, it includes:

[0038] When the threat level is low, trigger a warning signal and display a reminder box to alert the monitoring personnel;

[0039] When the threat level is medium, activate the electronic fence and send a forced departure signal in combination with an audible and visual alarm;

[0040] When the threat level is high, the UAV is pre-warned and defended through the linkage defense system;

[0041] The linkage defense system starts the directional interference device to block the UAV control signal and controls the capture net for physical interception.

[0042] Preferably, optimizing the UAV detection accuracy of the dynamic threat assessment result through the multi-modal feature fusion technology includes:

[0043] Screening UAV targets by analyzing and comparing the optical flow characteristics of UAVs and birds through optical flow characteristic analysis;

[0044] Optical flow calculation formula:

[0045]

[0046] In the formula, is the change rate of the image in the horizontal direction, is the change rate of the image in the vertical direction, is the change rate of the image pixel value over time, u is the movement in the horizontal direction, and v is the movement in the vertical direction;

[0047] Extracting visual features at different scales through the SIFT or NGF algorithm;

[0048] SIFT feature extraction formula:

[0049]

[0050] In the formula, I(x,y,σ) is the scale space of the image, and (x,y,σ) is the position of the feature point;

[0051] NGF gradient extraction formula:

[0052]

[0053] In the formula, I(x,y) is the pixel value of the image at the coordinates (x,y), and are the gradients of the image in the x and y directions respectively, and NGF(x,y) is the total gradient of the image at the coordinate point (x,y);

[0054] Filtering misdetected targets by analyzing the characteristics of the broken line trajectory of the UAV and the curve trajectory of the bird;

[0055] Optimizing the detection accuracy through a three-level feature fusion mechanism.

[0056] Preferably, based on the security level assessment result, combined with meteorological data and digital elevation model, the sensitivity information of the monitoring area is updated in real time, including:

[0057] Establish a sensitive database containing the spatial coordinates of no-fly zones and restricted-fly zones and their corresponding weight coefficients;

[0058] Realize dynamic update of environmental parameters through the fusion of digital elevation model and meteorological data;

[0059] Adopt the Kalman filtering algorithm to achieve the optimal estimation of environmental parameters;

[0060] Combined with the security level assessment result, dynamically adjust the sensitivity of the real-time monitoring area.

[0061] Preferably, the method adopts a "device-edge-cloud" collaborative computing mode, including:

[0062] Edge computing nodes are used to deploy a lightweight YOLOv5 model for preliminary screening to achieve millisecond-level response;

[0063] Cloud servers are used to perform fine classification and trajectory analysis to provide global situation awareness;

[0064] Data collaboration mechanism is used to realize real-time synchronization and task distribution of device-edge-cloud data through edge gateways;

[0065] Adopt MapReduce or task partitioning algorithm to allocate tasks to edge computing nodes and cloud servers for computing:

[0066]

[0067] In the formula, f(s) is the global task, and f m (s) is the task of each computing node.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] The present invention synchronously collects multi-modal visual data of unmanned aerial vehicles through a multi-camera array, combines a dynamic threat assessment model and weighted Euclidean distance to quantify the threat level, significantly improving the anti-interference ability of unmanned aerial vehicle detection; and adopts optimized visual data preprocessing and feature extraction technologies, such as denoising, enhancement, hybrid difference algorithm and optical flow analysis, etc., effectively improving the detection accuracy; at the same time, through multi-modal feature fusion and dynamic environmental parameter modeling, the sensitivity is updated in real time, further enhancing the adaptability. The present invention accurately identifies low-altitude targets and triggers corresponding security measures according to different threat levels to ensure the security of the monitoring area. In addition, adopting the "device-edge-cloud" collaborative computing mode improves the response speed and computing efficiency, realizing accurate and efficient unmanned aerial vehicle intrusion detection and security assessment. Description of the Drawings

[0070] Figure 1 This is a flowchart of the method for drone intrusion detection and security level assessment based on visible light vision of the present invention. Specific implementation manners

[0071] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0072] Embodiment 1:

[0073] Please refer to Figure 1 As shown, a method for drone intrusion detection and security level assessment based on visible light vision includes:

[0074] Using a multi-camera array to synchronously collect multi-modal visual data of the drone;

[0075] The multi-camera array adopts a distributed architecture and maintains a stable proportion of the target in the picture by dynamically adjusting the camera focal length;

[0076] The multi-modal visual data includes the position information, speed information, trajectory deviation information, and environmental parameter information of the drone.

[0077] Preprocessing and feature extraction are performed on the collected visual data;

[0078] The preprocessing includes denoising processing and enhancement processing;

[0079] The denoising processing uses a spatio-temporal domain joint filtering algorithm to remove the noise in the visual data;

[0080] The enhancement processing uses an adaptive enhancement algorithm based on the Retinex theory to enhance the area where the target is located;

[0081] The feature extraction uses an improved YOLOv5 network to extract the key feature information of the drone;

[0082] An attention mechanism is introduced to optimize the sensitivity of the YOLOv5 network to details.

[0083] Furthermore, by combining the denoising processing, enhancement processing, and feature extraction of the improved YOLOv5 network, the performance of drone target recognition is significantly improved. These measures together enhance the detection and recognition ability of the target in a complex environment.

[0084] Using a hybrid differential algorithm to solve the problems of sudden light changes and motion speed sensitivity;

[0085] The hybrid difference algorithm combines the background difference method and the three-frame difference method;

[0086] Among them, the difference between the current frame and the background model is calculated by the background difference method to generate a background mask;

[0087] The difference between the current frame and the previous frame is calculated by the three-frame difference method to generate an inter-frame mask;

[0088] The background mask and the inter-frame mask are fused to obtain a moving target mask;

[0089] Based on the moving target mask, the background model is updated by the exponential weighted average method;

[0090] The histogram contrast algorithm is introduced to extract the significant region and narrow the range of candidate targets.

[0091] Furthermore, the hybrid difference algorithm effectively solves the problems of sudden light changes and motion speed sensitivity. By combining the background difference method and the three-frame difference method, it can not only accurately extract moving targets under changing lighting conditions but also reduce errors caused by differences in motion speed. This method improves the system's ability to detect moving targets in complex environments.

[0092] A dynamic threat assessment model is constructed based on the key feature information of the UAV, and the weighted Euclidean distance is used to quantify the threat level of the UAV in real time;

[0093] The dynamic threat assessment model is constructed based on the multi-dimensional parameter distance function;

[0094] The multi-dimensional parameters include the position, speed, trajectory deviation degree of the UAV and environmental parameters;

[0095] The distance function adopts the weighted Euclidean distance formula to quantify the threat level of the UAV in real time;

[0096] When constructing the model, the spatial coordinates and weight coefficients of the no-fly zone and restricted-fly zone in the monitoring area are referred to.

[0097] The safety level assessment results are divided into three levels: low, medium, and high. After generating the safety level assessment results, it also includes:

[0098] When the threat level is low, a warning signal is triggered and a reminder is displayed to the monitoring personnel through a bounding box;

[0099] When the threat level is medium, an electronic fence is activated and a forced departure signal is sent in combination with an audible and visual alarm;

[0100] When the threat level is high, the UAV is warned and defended through the linkage defense system;

[0101] The linkage defense system activates the directional jamming device to block the UAV control signal and controls the capture net for physical interception.

[0102] Furthermore, by constructing a dynamic threat assessment model, combining multi-dimensional parameters of the UAV (such as position, speed, trajectory deviation, etc.) with environmental factors, the threat level of the UAV is quantified in real time, enhancing the intelligence and accuracy of the UAV monitoring and defense system. Through this multi-level defense mechanism, different threat scenarios can be effectively addressed, ensuring the efficiency and accuracy of UAV safety management.

[0103] Optimize the UAV detection accuracy of the dynamic threat assessment model through multi-modal feature fusion technology;

[0104] Screen UAV targets by analyzing and comparing the optical flow characteristics of UAVs and birds through optical flow characteristics analysis;

[0105] Extract visual features at different scales through SIFT or NGF algorithms;

[0106] Filter misdetected targets by analyzing the characteristics of the broken-line trajectory of UAVs and the curve trajectory of birds;

[0107] Achieve detection accuracy optimization through a three-level feature fusion mechanism;

[0108] Use the optimized dynamic threat assessment model to evaluate the safety level of the current UAV and generate a safety level assessment result.

[0109] Furthermore, a safety level assessment result is generated through the dynamic threat assessment model, and the UAV detection accuracy is optimized by combining multi-modal feature fusion technology. The optical flow characteristics analysis is used to distinguish UAVs from birds, and different-scale visual features are extracted through SIFT or NGF algorithms. Further, misdetected targets are filtered through trajectory analysis. The three-level feature fusion mechanism significantly improves the detection accuracy, reduces the risk of false alarms and missed detections, and effectively improves the reliability and precision of UAV identification and threat assessment.

[0110] According to the safety level assessment result, combined with meteorological data and digital elevation model, the sensitivity information of the monitoring area is updated in real time;

[0111] Establish a sensitive database containing the spatial coordinates of no-fly zones and restricted-fly zones and the corresponding weight coefficients;

[0112] Realize dynamic update of environmental parameters through digital elevation model fusion of meteorological data;

[0113] Adopt the Kalman filter algorithm to achieve the optimal estimation of environmental parameters;

[0114] The modeling process is dynamically adjusted according to the sensitivity of the real-time monitoring area.

[0115] Furthermore, by combining meteorological data with the digital elevation model, the sensitivity information of the monitored area is updated in real time, and a sensitive database containing no-fly zones, restricted-fly zones, and their weight coefficients is established. Through the fusion of the digital elevation model and meteorological data, the environmental parameters are dynamically adjusted, and the Kalman filtering algorithm is used to achieve the optimal estimation of the environmental parameters, thus ensuring the real-time and accurate update of the sensitivity of the monitored area. This process of dynamic modeling and real-time adjustment significantly improves the flexibility and accuracy of area monitoring, and effectively enhances the ability to ensure the management of unmanned aerial vehicles and flight safety.

[0116] This method adopts a "terminal-edge-cloud" collaborative computing mode, including:

[0117] Edge computing nodes, which are used to deploy the lightweight YOLOv5 model for preliminary screening to achieve millisecond-level response;

[0118] Cloud servers, which are used to perform fine classification and trajectory analysis to provide global situation awareness;

[0119] A data collaboration mechanism, which is used to realize real-time synchronization and task distribution of terminal-edge-cloud data through an edge gateway;

[0120] The MapReduce or task partitioning algorithm is used to allocate tasks to edge computing nodes and cloud servers for computing.

[0121] Example 2:

[0122] Application example: Security level assessment of a certain international airport detecting drone intrusion

[0123] There have been frequent incidents of unknown drones breaking into the airspace around a certain international airport. Due to ground clutter interference, traditional radar systems are difficult to effectively identify low-altitude micro drones (RCS < 0.5m 2 ), and the detection range of infrared thermal imaging is less than 50 meters in morning fog weather, and the false alarm rate of radio monitoring is as high as 42%. To improve the security efficiency, the airport introduces a drone intrusion detection and security level assessment method based on visible light vision.

[0124] I. Scheme implementation

[0125] (1) Multimodal perception network

[0126] Deploy 48 4K cameras along the perimeter of the runway to form a ring array. The cost of a single camera is only ￥480. In cooperation with the Hikvision DS-2CD3T46WD-I3 device, it can achieve full coverage of the airspace at an altitude of 15 - 200 meters.

[0127] Integrate the hybrid differential algorithm to maintain a detection accuracy of 92% in scenarios with sudden light changes (> 1000 lux change).

[0128] (2) Edge computing architecture

[0129] For every 3 cameras, 1 Jetson Xavier NX computing node (￥3,200 per unit) is configured, and a lightweight YOLOv5 model is deployed to achieve real-time processing at 12 ms / frame.

[0130] A dynamic threat assessment model is adopted, which integrates three-dimensional features of position, speed, and trajectory deviation, and quantifies the threat level through the weighted Euclidean distance algorithm.

[0131] (3) Anti-interference enhancement strategy

[0132] Develop an optical flow feature discrimination module, establish a database of the movement patterns of drones and birds, and achieve a 98.2% filtering of false detections of birds.

[0133] Combined with the digital elevation model and meteorological data, dynamically adjust the detection parameters, and the detection distance is increased by 180% compared with the infrared system in haze weather.

[0134] (4) Performance comparison

[0135]

[0136] (5) Safety decision support

[0137] Automatically trigger a three-level response mechanism according to the threat level:

[0138] Low-level threat: Mark an orange warning box on the monitoring interface and synchronously push it to the handheld terminal of security personnel

[0139] Medium-level threat: Activate the directional sound wave dispersion system and emit a 120 dB warning audio. High-level threat: Link the anti-drone defense system and emit directional electromagnetic waves to implement a forced landing. II. Benefit analysis

[0140] This solution reduces the airport security cost by 95%, improves the detection accuracy by 27%, and still maintains a stable detection rate of 93% in an environment with an electromagnetic interference intensity > 60 dBm. After actual measurement, it successfully detected a DJI Mavic 3 drone 18 meters above the ground, while the traditional radar did not trigger an alarm, demonstrating significant advantages in a complex electromagnetic environment.

[0141] Example 3:

[0142] An embodiment of the present invention also provides a computer-readable storage medium. A program of a method for drone intrusion detection and security level assessment based on visible light vision as described in any one of the above is stored on the computer-readable storage medium. When the program is executed by a processor, it implements each process of the above-described embodiment of the assessment method and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0143] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0144] In the attached drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments of the present invention are involved. Other structures can refer to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0145] The flowcharts shown in the drawings are only illustrative and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0146] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting drone intrusion and evaluating security level based on visible light vision, characterized in that, Including: Utilize a multi-camera array to synchronously collect multi-modal visual data of the UAV, preprocess and extract features from the collected visual data to obtain key feature information of the UAV; Construct a dynamic threat assessment model based on the key feature information of the UAV, and use weighted Euclidean distance to quantify the threat level of the UAV in real time; Optimize the UAV detection accuracy of the dynamic threat assessment model through multi-modal feature fusion technology; Use the optimized dynamic threat assessment model to evaluate the safety level of the current UAV and generate a safety level assessment result; According to the safety level assessment result, combine meteorological data and digital elevation model to update the sensitivity information of the monitoring area in real time.

2. The method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 1, wherein: The multi-camera array adopts a distributed architecture and maintains a stable proportion of the target in the picture by dynamically adjusting the camera focal length; the multi-modal visual data includes the position information, speed information, trajectory deviation information and environmental parameter information of the UAV.

3. A method for drone intrusion detection and security level assessment based on visible light vision according to claim 2, characterized in that: The preprocessing and feature extraction of the collected visual data to obtain key feature information of the UAV includes: Adopt a spatio-temporal domain joint filtering algorithm to remove noise in the visual data. The formula of the spatio-temporal domain joint filtering algorithm is as follows: Wherein, E′ is the denoised image, K is the normalization constant, E a,b and G a,b are the image pixel value and the filter kernel value respectively, and S a,b is the spatio-temporal domain weighting coefficient; Adopt an adaptive enhancement algorithm based on the Retinex theory to enhance the area where the target is located; Adopt an improved YOLOv5 network to extract the feature information of the UAV, introduce an attention mechanism to optimize the sensitivity of the YOLOv5 network to details, and obtain the key feature information of the UAV: In the formula, A(x,y) represents the attention weight, W is the weight matrix, and F(x,y) is the image feature.

4. A method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 3, characterized in that: The preprocessing and feature extraction of the collected visual data to obtain key feature information of the UAV also includes: Use a hybrid difference algorithm to solve the problems of sudden light change and motion speed sensitivity; The hybrid difference algorithm combines the background difference method and the three-frame difference method; Among them, calculate the difference between the current frame and the background model through the background difference method to generate a background mask; Use the three-frame difference method to calculate the difference between the current frame and the previous frame to generate an inter-frame mask; Fuse the background mask and the inter-frame mask to obtain a moving target mask; Based on the moving target mask, use the exponential weighted average method to update the background model; Introduce a histogram contrast algorithm to extract the significant area and narrow the range of candidate targets; Based on the narrowed range of candidate targets, adopt an improved YOLOv5 network to extract the key feature information of the UAV.

5. The method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 4, wherein: The constructing a dynamic threat assessment model based on the key feature information of the UAV and using weighted Euclidean distance to quantify the threat level of the UAV in real time includes: Adopt a multi-dimensional parameter distance function to construct the dynamic threat assessment model; the multi-dimensional parameter distance function uses the weighted Euclidean distance formula to quantify the threat level of the UAV in real time based on the position, speed, trajectory deviation and environmental parameters of the UAV; where w r is the weight factor of the r-th parameter, h r is the parameter value collected in real time, h r() is the preset safety threshold, and n is the total number of parameters.

6. The method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 5, wherein: The safety level assessment result is divided into three levels: low, medium and high; After generating the safety level assessment result, it also includes: When the threat level is low, trigger a warning signal and display a reminder to the monitoring personnel through a marking box; When the threat level is medium, activate the electronic fence and send a forced departure signal in combination with an audible and visual alarm; When the threat level is high, the UAV is pre-warned and defended through the linkage defense system; The linkage defense system starts the directional interference device to block the UAV control signal and controls the capture net for physical interception.

7. A method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 6, characterized in that: Optimizing the UAV detection accuracy of the dynamic threat assessment model through multi-modal feature fusion technology includes: Screening UAV targets by analyzing and comparing the optical flow characteristics of UAVs and birds through optical flow characteristics; Optical flow calculation formula: Wherein, is the change rate of the image in the horizontal direction, is the change rate of the image in the vertical direction, is the change rate of the image pixel value over time, u is the motion in the horizontal direction, and v is the motion in the vertical direction; Extracting visual features at different scales through the SIFT or NGF algorithm; SIFT feature extraction formula: In the formula, I(x, y, σ) is the scale space of the image, and (x, y, σ) is the position of the feature point; NGF gradient extraction formula: where I(x, y) is the pixel value of the image at the coordinate (x, y), and are the gradients of the image in the x and y directions respectively, and NGF(x, y) is the total gradient of the image at the coordinate point (x, y); Filtering misdetected targets by analyzing the broken line trajectory of the UAV and the curve trajectory characteristics of birds; Optimizing the detection accuracy through a three-level feature fusion mechanism.

8. The method for detecting drone intrusion and evaluating security level based on visible light vision according to claim 7, characterized in that: Updating the sensitivity information of the monitoring area in real time according to the security level assessment result combined with meteorological data and digital elevation model, including: Establishing a sensitive database containing the spatial coordinates of no-fly zones and restricted-fly zones and the corresponding weight coefficients; Realizing dynamic update of environmental parameters through digital elevation model fusion of meteorological data; Adopting the Kalman filter algorithm to achieve the optimal estimation of environmental parameters; Combining the security level assessment result to dynamically adjust the sensitivity of the real-time monitoring area.

9. A method for drone intrusion detection and security level assessment based on visible light vision according to claim 8, characterized in that, The method adopts a "terminal-edge-cloud" collaborative computing mode, including: Edge computing nodes, used to deploy the lightweight YOLOv5 model for preliminary screening to achieve millisecond-level response; Cloud servers, used to perform fine classification and trajectory analysis to provide global situation awareness; Data collaboration mechanism, used to realize real-time synchronization and task distribution of terminal-edge-cloud data through the edge gateway; Adopting MapReduce or task division algorithm to allocate tasks to edge computing nodes and cloud servers for calculation: where f(s) is the global task, and f m (s) is the task of each computing node.

Images