Airport boundary grading early warning linkage method and system based on fuzzy logic reasoning

Abstract

The application belongs to the technical field of airport perimeter security monitoring, and discloses an airport perimeter hierarchical early warning linkage method and system based on fuzzy logic reasoning. The method generates infrared and visible light panoramic images through panoramic scanning of the airport perimeter, detects and identifies the intrusion targets in the perimeter defense area in the infrared and visible light panoramic images through a multi-modal target detection algorithm based on deep learning, divides the intrusion target warning levels through a fuzzy logic algorithm, judges the priority of the intrusion target warning levels, uses an OPENCV algorithm library to mark the trajectories of multiple coordinate points closest to the intrusion target in the perimeter panoramic base map, displays the motion trajectory of the current intrusion target, and judges the intention of the intrusion target. The application can adapt to different airport environments through the self-optimization capability of the rule base, reduce the artificial patrol load while ensuring high reliability, and enhance the intelligent level and safety and control efficiency of the perimeter security system.

Description

Technical Field

[0001] This invention belongs to the field of airport perimeter security monitoring technology, and particularly relates to a hierarchical early warning and linkage method and system for airport perimeter based on fuzzy logic reasoning. Background Technology

[0002] In recent years, with the rapid development of the air transport industry, airport perimeter security systems have faced increasingly complex security challenges. Security risks exist at airports.

[0003] Traditional airport perimeter systems often employ single technologies such as infrared beam detectors, electronic fences, and vibration-based fiber optics. However, these technologies have significant drawbacks, being highly susceptible to interference from severe weather conditions such as wind, snow, and heavy rain, resulting in consistently high false alarm rates and failing to meet the ever-increasing demands of airports for perimeter security systems. For example, a perimeter system based on traditional fiber optic sensors can experience hundreds of false alarms daily during typhoon weather.

[0004] Furthermore, relevant industry standards clearly stipulate that airport perimeter alarm systems must possess tiered early warning capabilities, accurately identifying different types of intrusion behavior and taking corresponding preventative measures based on the degree of danger. However, traditional airport perimeter systems typically lack alarm tiering capabilities. When multiple targets exist within the perimeter area, they cannot effectively and intelligently tier intrusion targets to assist perimeter staff in accurately locating the highest-level target. Even systems with alarm tiering capabilities mostly rely on fixed thresholds and preset fixed rules, resulting in rigid static tiering algorithms. For example, the system often defines a defense zone within 5 meters of the perimeter, failing to effectively identify pre-intrusion behaviors such as throwing or loitering outside 5 meters, thus failing to meet the actual needs of dynamic airport security. Therefore, a more intelligent and effective method and linkage system for achieving intelligent tiered early warning in airport perimeter systems is needed. Summary of the Invention

[0005] To overcome the problems existing in related technologies, the embodiments disclosed in this invention provide a method and system for hierarchical early warning linkage of airport perimeter based on fuzzy logic reasoning. The purpose of this invention is to solve the technical problems of high false alarm rates, lack of intrusion risk level classification, and rigid static alarm classification in existing technologies, which rely on single detection technologies.

[0006] The technical solution is as follows: A hierarchical early warning and linkage method for airport perimeter based on fuzzy logic reasoning, comprising the following steps:

[0007] S1, panoramic scanning of the airport perimeter generates infrared and visible light panoramic images, and sends the generated infrared and visible light panoramic image information to the outside world through the MQTT network protocol;

[0008] S2 receives infrared and visible light panoramic image data, and uses a deep learning-based multimodal target detection algorithm to detect and identify intrusion targets in the perimeter defense zone area of ​​the infrared and visible light panoramic images; it pushes the early warning information of the detected and identified intrusion targets, which includes the coordinates of the intrusion targets and the intrusion feature information of the intrusion targets;

[0009] S3 receives and updates the transmitted infrared and visible light panoramic images in real time as the perimeter panoramic base map. It uses a fuzzy logic algorithm to classify the intrusion target warning information and determine the priority of the intrusion target warning level. It then pushes the intrusion target warning level after priority determination via the HTTP protocol.

[0010] S4, based on the priority judgment of the push notification and the warning level of the intrusion target, uses the OPENCV algorithm library to mark the trajectory of multiple nearest coordinate points of the intrusion target in the perimeter panoramic base map, displays the current movement trajectory of the intrusion target, and judges the intrusion target's intent.

[0011] In step S1, the airport perimeter is scanned using a dual-band panoramic monitoring subsystem to generate infrared and visible light panoramic images. The dual-band panoramic monitoring subsystem includes multiple panoramic scanning devices, each installed at intervals, and mounted simultaneously with a PTZ camera on the horizontal arm of a pole, which is installed inside the airport perimeter. Through continuous horizontal 360-degree rotational scanning, each device generates a 360-degree panoramic image at fixed time intervals. The specific steps are as follows:

[0012] Step 1: Establish a mechanism model for reducing interference light perturbation in grayscale images and determine the optimization framework;

[0013] Step 2: Construct data items for modeling light disturbances;

[0014] Step 3: Select image prior terms and combine them with data terms to construct a non-optimal feedback model for grayscale images affected by interference light.

[0015] Step four: Obtain a clear panoramic image using the iterative reweighted second-order iterative optimization method.

[0016] In step one, a mechanism model for mitigating interference light perturbation in grayscale images is established, and an optimization framework is determined, including:

[0017] (1) Establish a nonlinear attenuation model for interference light disturbance in grayscale images, the expression of which is:

[0018]

[0019] In the formula, Let f be the vector of the blurred image that has not been contaminated by grayscale, D be the known blur kernel, u be the vector of the original clear image, f be the vector of the final weakened image, and W be the process of the image being contaminated by light disturbance.

[0020] (2) Based on the weakening model and the maximum a posteriori probability framework, data fidelity and image prior are combined into a unified image restoration framework. The non-optimal feedback problem corresponds to the optimization problem, and its expression is:

[0021]

[0022] In the formula, To unify the image vectors in the image restoration framework, C(u|f) is the posterior probability;

[0023] According to Bayes' theorem, we get:

[0024] C(u|f)∝C(f|u)C(u)

[0025] In the formula, C(f|u) is the likelihood probability and C(u) is the prior probability;

[0026] Combining the two formulas above, the equivalent transformation is:

[0027]

[0028] In the formula, The image vector after the equivalent transformation;

[0029] The following optimized framework was obtained:

[0030]

[0031] In the formula, ψ(Du-f) is the data item to be constructed. The priors for the image to be designed;

[0032] Step two involves constructing data items for modeling light disturbances, specifically including:

[0033] (1) Select the following nonlinear function to eliminate the influence of perturbation grayscale:

[0034]

[0035] In the formula, h and l are both function parameters. h controls the degree of nonlinearity of the function at the cutoff point, l controls the cutoff point, D(u) is the fuzzy kernel after removing the influence of perturbation grayscale, and e is the Reynolds number.

[0036] (2) Construct the following data terms by combining the nonlinear functions of the L1 norm:

[0037]

[0038] In the formula, i is the image pixel index, f is the vector of the final weakened image, S(u) is the data item of the image vector, and Q is a nonlinear function;

[0039] In step three, a non-optimal feedback model for the grayscale image with interference light perturbation is constructed by combining the data items, including:

[0040] The expression for the prior term, constructed based on the sparse prior of the image gradient, is as follows:

[0041]

[0042] In the formula, L w Let w be the sparse prior value of the image gradient at the w-th image node. For image gradient sparse prior non-optimal feedback, L v The sparse prior value of the image gradient for the v-th image node;

[0043] In step four, the iterative reweighted second-order iterative optimization method includes:

[0044] The non-optimization feedback model is transformed into a weighted second-order iterative optimization method as shown in the following equation:

[0045]

[0046] In the formula, λ u For the weight of the regularization term, For weighted second-order iterative optimization, A w Let D be the diagonal matrix of the weights of the sparse prior values ​​of the image gradient in the deformation of the w-th image node. w Let A be the blur kernel in the deformation of the w-th image node. v Let D be the diagonal matrix of the weights of the sparse prior values ​​of the image gradient in the deformation of the v-th image node. v Let f be the blur kernel in the deformation of the v-th image node.

[0047] In step S2, infrared and visible light panoramic image data are received, including:

[0048] By subscribing to MQTT events, the AI ​​intelligent analysis center receives infrared and visible light panoramic images sent by the dual-band panoramic monitoring subsystem. Through a deep learning-based multimodal target detection algorithm, it detects and identifies intrusion targets in the perimeter defense area of ​​the infrared and visible light panoramic images, and pushes the early warning information of the detected and identified intrusion targets to the intelligent alarm classification and processing center.

[0049] The intrusion target detection includes target type, target behavior, target distance, and target dwell time. Target type includes characteristics of people, vehicles, animals, and drones. Target behavior includes characteristics of being stationary, loitering, climbing, and throwing. The target distance is the vertical distance between the target and the virtual boundary in the panoramic image; the closer the distance, the higher the risk of the intrusion target. The target dwell time is the time the intrusion target spends within the boundary defense zone.

[0050] The deep learning-based multimodal target detection algorithm includes: first, generating an event code based on the UUID standard and attaching spatiotemporal coordinate labels; then, extracting a multispectral fusion feature vector, including RGB spectrum distribution, infrared radiation intensity, and morphological contour parameters; simultaneously activating a time-series analysis engine, combining it with the device's historical image database to reconstruct motion trajectories and match behavior patterns, and establishing a three-dimensional spatial motion situation model of the target; finally, outputting structured feature dataset parameters, including target type, target behavior, target distance, and target dwell time.

[0051] Establishing a three-dimensional spatial motion situation model of the target includes:

[0052]

[0053] In the formula, X, T, K, t represent the target type motion value, target behavior motion trend motion value, target distance, and target dwell time, respectively; G represents the target motion matching parameter; J represents the multispectral fusion feature sparsity; z represents the vertical target motion region; F represents the multispectral fusion feature vector; x represents the horizontal target motion region; y represents the vertical target motion region; and H represents the target motion region. z I is the vertical target tendency motion coefficient, and I is the radius of the target's trajectory.

[0054] In step S3, the intrusion target warning level is classified using a fuzzy logic algorithm, and the priority of the intrusion target warning level is determined, including:

[0055] S301, Design of fuzzification of intrusion target feature variables;

[0056] S302, Design of a fuzzy rule base for intrusion alarm;

[0057] S303, Fuzzy Reasoning and Defuzzification of Intrusion Alarm Features;

[0058] S304, the intelligent alarm classification processing center saves the intrusion target warning level obtained by fuzzy algorithm reasoning to the database, and pushes the intrusion target coordinates and intrusion target warning level to the linkage control center via HTTP protocol.

[0059] In step S301, the design of fuzzification of the intrusion target feature variables includes:

[0060] The expression for the membership function of the target type is:

[0061]

[0062] In the formula, μ type ω represents the membership degree of the target type, ranging from [0,1]; type represents the target type risk weight, ranging from [0,1]; reliability represents the confidence level of the target behavior.

[0063] The membership function expression for the target behavior is:

[0064]

[0065] In the formula, μ behavior ω represents the membership degree of the target behavior, ranging from [0,1]. behavior The target behavior risk weight is defined in the range [0,1].

[0066] The target distance membership function expression is:

[0067]

[0068] In the formula, μ distance d is the membership degree of the target distance, ranging from [0,1]; d is the vertical distance of the target distance from the boundary wall, m is the center point, m=5; σ is the standard deviation, σ=5;

[0069] The membership function expression for the target residence time is:

[0070]

[0071] In the formula, μ time The membership degree is defined by the target residence time, ranging from [0,1]; k is the slope coefficient, and T mid For the midpoint time, T crit This is the critical time.

[0072] In step S302, the design of the intrusion alarm fuzzy rule base includes:

[0073] The design of the fuzzy rule base satisfies the following rules:

[0074] (1) The fuzzy rule base should cover all input combination states;

[0075] (2) The four characteristic variables of airport perimeter intrusion targets—target type, target behavior, target distance, and target dwell time—are divided into full spatial categories.

[0076] (3) Natural language description rules are adopted, namely the IF-THEN structure, and the expression form is:

[0077] IF(μ type ∈ Ltype AND (μ) behavior ∈L behavior AND (μ) distance ∈L distance AND (μ) ditance

[0078] ∈L time THEN Warning Level = C m

[0079] Where L is the fuzzy level of the feature variable, and C m The rule outputs the ultra-high / high / medium / low warning levels; L type For the fuzzy level of the feature variable of the target type, L behavior For the fuzzy level of the characteristic variable of the target behavior, L distance For the fuzzy level of the feature variable of target distance, L time Fuzzy ranking of the characteristic variable of target residence time;

[0080] (4) The fuzzy rule strength calculation should follow the AND calculation method to calculate the rule strength information that the intrusion target characteristic variables conform to. The expression is:

[0081] μ rulei =min(μ type ,μ behavior ,μ distance ,μ time )×ω i

[0082] In the formula, μ rulei To adapt the activation strength of the i-th rule to the intrusion target, ω i Let be the rule weight of the i-th rule.

[0083] In step S303, the fuzzy inference and defuzzification of intrusion alarm features include:

[0084] Based on fuzzy feature variables and fuzzy rule base, and taking the feature variables of the intrusion target as input, the system performs reasoning and finally derives the warning level of the intrusion target by following the steps of feature variable fuzzification, rule strength evaluation, rule consequent aggregation and defuzzification.

[0085] The rule strength evaluation is based on a predefined fuzzy rule library. For each rule antecedent, logical operations are performed according to the fuzzy rule strength calculation formula to obtain the activation strength of the rule consequent.

[0086] The rule consequent aggregation merges the fuzzy sets of all activated rule consequents into a comprehensive output fuzzy set, using the maximum value method (MAX) according to arg max(μ).rulei The results of multiple rules are superimposed and merged in a manner that allows for the aggregation and merging of μ. rule =max(μ rule1 ,μ rule2 ,μ rule3 …μ rulen The alarm rule strength value is obtained, and then dynamically divided according to the activation strength threshold maintained by the system. Common dynamic optimization methods include manual configuration, data-driven optimization, and real-time adaptation.

[0087] Manual configuration includes manually adjusting threshold ranges through a management interface; data-driven optimization includes automatically fitting the optimal threshold based on historical intrusion event data and using machine learning to adapt to the environmental differences of different airports.

[0088] Real-time adaptive features include: dynamically scaling the threshold range by combining real-time weather and passenger flow data; the warning level activation intensity range includes four levels: ultra-high [0.8,1], high [0.6,0.8], medium [0.3,0.6], and low [0.0,0.3].

[0089] In step S4, the OpenCV algorithm library is used to annotate the trajectory of multiple nearest coordinate points of the intrusion target in the perimeter panoramic base map, displaying the current movement trajectory of the intrusion target and determining the intrusion target's intent, including:

[0090] The joint control center can automatically identify and lock onto the most threatening target based on the intrusion target warning level; by mapping the intrusion target coordinates in the perimeter panoramic base map to the horizontal rotation angle P, pitch angle T, and zoom magnification Z of the PTZ camera in real time, it can track the intrusion target; at the same time, the joint control center obtains the PTZ camera information bound to the current perimeter scanning equipment from the database, and constructs a three-dimensional spatial coordinate transformation model based on the mapping relationship between the perimeter panoramic base map coordinate system and the PTZ camera physical coordinate system, automatically controlling the PTZ camera to turn towards the intrusion target direction for video tracking and confirmation.

[0091] Furthermore, the joint control center calculates the PTZ parameters of the PTZ camera using the coordinates (PositionX, PositionY) of the intruding target, the vertical field of view (ForVer) and the pitch angle (DangleVer) of the peripheral scanning equipment; the specific mapping process is as follows:

[0092] The horizontal angle P is calculated using a polar coordinate mapping algorithm, converting the target pixel coordinates (PositionX / ImgWidth) into the PTZ camera's horizontal rotation angle.

[0093]

[0094] In the formula, ImgWidth and ImgHeight are the pixel width and height of the panoramic image, respectively, and θoffset The initial angle calibration offset between the peripheral scanning equipment and the PTZ camera is pre-measured and stored through the equipment calibration program. This offset is applied when the peripheral scanning equipment and the PTZ camera are at the same angle θ. offset =0;

[0095] The vertical pitch T is calculated using a trigonometric algorithm, combining the PTZ camera's vertical field of view ForVer and pitch angle DangleVer.

[0096]

[0097] In the formula, the negative sign indicates the engineering convention that the downward direction of the PTZ camera is the positive direction;

[0098] The zoom factor Z is:

[0099]

[0100] In the formula, Z min Z max These are the minimum and maximum zoom levels preset for the PTZ camera, respectively.

[0101] Another objective of this invention is to provide an airport perimeter hierarchical early warning and linkage system based on fuzzy logic reasoning. This system implements the aforementioned airport perimeter hierarchical early warning and linkage method based on fuzzy logic reasoning. The system includes:

[0102] The dual-band panoramic monitoring subsystem is used to scan the airport perimeter to generate infrared and visible light panoramic images, and to send the generated infrared and visible light panoramic image information to the outside world via the MQTT network protocol.

[0103] The AI ​​intelligent analysis center receives infrared and visible light panoramic image data and uses a deep learning-based multimodal target detection algorithm to detect and identify intrusion targets in the perimeter defense zone of the infrared and visible light panoramic images. It then pushes out early warning information for the detected and identified intrusion targets, which includes the coordinates of the intrusion target and its intrusion characteristics.

[0104] The intelligent alarm classification and processing center is used to receive and update infrared and visible light panoramic images in real time as the perimeter panoramic base map. It classifies the intrusion target warning information through fuzzy logic algorithm, judges the priority of the intrusion target warning level, and pushes the intrusion target warning level after priority judgment through HTTP protocol.

[0105] The linkage control center is used to determine the priority of the push notifications and the warning level of the intrusion target. It uses the OPENCV algorithm library to mark the trajectory of multiple nearest coordinate points of the intrusion target on the perimeter panoramic base map, display the current movement trajectory of the intrusion target, and determine the intrusion target's intent.

[0106] Combining all the above technical solutions, the beneficial effects of this invention are as follows: By combining panoramic images generated by a 360-degree perimeter scanning device, this invention utilizes AI to intelligently identify the characteristics and behaviors of intruders, and focuses on using fuzzy logic algorithms for reasoning calculations to achieve intelligent hierarchical early warning and PTZ camera linkage tracking and confirmation functions for perimeter intrusion targets, effectively reducing the false alarm rate and improving the accuracy and response speed of airport perimeter security alarms.

[0107] This invention constructs an intelligent multi-feature data fusion mechanism based on fuzzy logic reasoning, effectively reducing false alarms caused by environmental interference. The system employs an adaptive hierarchical early warning strategy, breaking through the limitations of traditional fixed thresholds. Through multi-dimensional feature analysis and real-time reasoning, it dynamically determines threat levels and links with PTZ cameras, audio-visual equipment, and other devices to form a hierarchical response closed loop, significantly improving emergency response efficiency and security accuracy. Through the self-optimization capability of the rule base, it adapts to different airport environments, reducing the workload of manual patrols while ensuring high reliability, thus enhancing the intelligence level and security control effectiveness of the perimeter security system. Attached Figure Description

[0108] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0109] Figure 1 This is a schematic diagram of the airport perimeter hierarchical early warning and linkage system based on fuzzy logic reasoning provided in an embodiment of the present invention;

[0110] Figure 2 This is a schematic diagram of the airport perimeter hierarchical early warning linkage method based on fuzzy logic reasoning provided in the embodiments of the present invention;

[0111] Figure 3 This is a flowchart of the fuzzy logic algorithm for logical reasoning of target features provided in this embodiment of the invention. Detailed Implementation

[0112] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0113] The innovation of this invention lies in the fact that it breaks through the traditional perimeter system's "single technology detection and fixed threshold grading" mode. It improves detection accuracy through multimodal perception (dual-band panoramic monitoring technology + deep learning multimodal target detection), achieves dynamic grading through fuzzy logic (multi-dimensional feature fuzzy modeling + full-space coverage fuzzy rule library), and optimizes response efficiency through intelligent linkage (panoramic base map trajectory visualization + three-dimensional coordinate mapping and PTZ camera linkage). This forms a complete intelligent security closed loop of "detection-grading-response", which significantly improves the security and control effectiveness and intelligence level of airport perimeters.

[0114] Example 1: The airport perimeter hierarchical early warning linkage method based on fuzzy logic reasoning provided in this embodiment of the invention performs AI intelligent analysis on the panoramic images scanned around the perimeter to identify characteristic factors such as the type, behavior, distance, and dwell time of the intruding target. Then, it calculates the feature membership degree through a fuzzy logic algorithm and performs hierarchical early warning management, and links the PTZ camera for video tracking. Figure 1 As shown.

[0115] Figure 2 This is the principle of the airport perimeter hierarchical early warning linkage method based on fuzzy logic reasoning provided in the embodiments of the present invention.

[0116] Specifically, the airport perimeter hierarchical early warning and linkage method based on fuzzy logic reasoning provided in this embodiment of the invention includes:

[0117] S1, panoramic scanning of the airport perimeter generates infrared and visible light panoramic images, and sends the generated infrared and visible light panoramic image information to the outside world through the MQTT network protocol;

[0118] For example, the dual-band panoramic monitoring subsystem includes multiple perimeter scanning monitoring devices capable of panoramic scanning. Each device is installed at a distance from the others and is mounted on the crossarm of a pole along with a PTZ camera. The pole is installed inside the airport perimeter. Through continuous horizontal 360-degree rotational scanning, each device generates a 360-degree panoramic image at fixed time intervals.

[0119] MQTT is a lightweight, publish-subscribe-based application layer messaging protocol suitable for resource-constrained devices and low-bandwidth, high-latency, or unstable network environments.

[0120] The weekly scanning monitoring equipment generates a 360-degree panoramic image at fixed time intervals, including:

[0121] Step 1: Establish a mechanism model for reducing interference light perturbation in grayscale images and determine the optimization framework;

[0122] Step 2: Construct data items for modeling light disturbances;

[0123] Step 3: Select image prior terms and combine them with data terms to construct a non-optimal feedback model for grayscale images affected by interference light perturbation.

[0124] Step four: Obtain a clear panoramic image using the iterative reweighted second-order iterative optimization method.

[0125] For example, in step one, establishing a blurred image attenuation mechanism model and determining the optimization framework specifically includes:

[0126] (1) Establish a nonlinear attenuation model for interference light disturbance in grayscale images, the expression of which is:

[0127]

[0128] In the formula, Let f be the vector of the blurred image that has not been contaminated by grayscale, D be the known blur kernel, u be the vector of the original clear image, f be the vector of the final weakened image, and W be the process of the image being contaminated by light disturbance.

[0129] (2) Based on the weakening model and the maximum a posteriori probability framework, data fidelity and image prior are combined into a unified image restoration framework. The non-optimal feedback problem corresponds to the optimization problem, and its expression is:

[0130]

[0131] In the formula, To unify the image vectors in the image restoration framework, C(u|f) is the posterior probability;

[0132] According to Bayes' theorem, we get:

[0133] C(u|f)∝C(f|u)C(u)

[0134] In the formula, C(f|u) is the likelihood probability and C(u) is the prior probability;

[0135] Combining the two formulas above, the equivalent transformation is:

[0136]

[0137] In the formula, The image vector after the equivalent transformation;

[0138] The following optimized framework was obtained:

[0139]

[0140] In the formula, ψ(Du-f) is the data item to be constructed. The image prior term to be designed is the regularization term, which constrains the panoramic image to conform to the manually designed statistical prior distribution, corresponding to the term -logC(u).

[0141] In step two, the method for constructing data items to model light disturbance specifically includes:

[0142] (1) Select the following nonlinear function to eliminate the influence of perturbation grayscale:

[0143]

[0144] In the formula, h and l are both function parameters. h controls the degree of nonlinearity of the function at the cutoff point, l controls the cutoff point, D(u) is the fuzzy kernel after removing the influence of perturbation grayscale, and e is the Reynolds number.

[0145] (2) Construct the following data terms by combining the nonlinear functions of the L1 norm:

[0146]

[0147] In the formula, i is the image pixel index, f is the vector of the final weakened image, S(u) is the data item of the image vector, and Q is a nonlinear function;

[0148] In step three, a non-optimal feedback model for the grayscale image with interference light perturbation is constructed by combining the data items, including:

[0149] The expression for the prior term, constructed based on the sparse prior of the image gradient, is as follows:

[0150]

[0151] In the formula, L w Let w be the sparse prior value of the image gradient at the w-th image node. For image gradient sparse prior non-optimal feedback, L v The sparse prior value of the image gradient for the v-th image node;

[0152] Step four, the iterative reweighted second-order iterative optimization method, includes:

[0153] The non-optimization feedback model is transformed into a weighted second-order iterative optimization method as shown in the following equation:

[0154]

[0155] In the formula, λ u For the weight of the regularization term, For weighted second-order iterative optimization, A w Let D be the diagonal matrix of the weights of the sparse prior values ​​of the image gradient in the deformation of the w-th image node. w Let A be the blur kernel in the deformation of the w-th image node.v Let D be the diagonal matrix of the weights of the sparse prior values ​​of the image gradient in the deformation of the v-th image node. v Let f be the blur kernel in the deformation of the v-th image node.

[0156] S2 receives infrared and visible light panoramic image data, and uses a deep learning-based multimodal target detection algorithm to detect and identify intrusion targets in the perimeter defense zone area of ​​the infrared and visible light panoramic images; it pushes the early warning information of the detected and identified intrusion targets, which includes the coordinates of the intrusion targets and the intrusion feature information of the intrusion targets;

[0157] The AI ​​Intelligent Analysis Center receives infrared and visible light panoramic images sent by the dual-band panoramic monitoring subsystem by subscribing to MQTT events.

[0158] For example, the deep learning-based multimodal target detection algorithm is an algorithm that can detect and analyze targets in infrared and visible light panoramic images. By detecting and analyzing, it obtains feature information such as target type, target behavior, target distance, and target dwell time of the intrusion target contained in the image, thereby providing basic target feature variables for subsequent fuzzy logic reasoning algorithms.

[0159] For example, the AI ​​intelligent analysis center can identify and set up virtual field perimeter walls in panoramic images, set up core defense zones 5 meters inside and outside the virtual perimeter walls, and set up early warning defense zones 20 meters inside and outside the virtual perimeter walls. Then, it detects the target type, target behavior, target distance, and target dwell time of all intrusion targets within the perimeter defense zones, and pushes the intrusion target coordinates and intrusion characteristic information of the intrusion target early warning information to the intelligent alarm classification and processing center.

[0160] For example, the intrusion target detection includes target type, target behavior, target distance, and target dwell time. The target type includes features such as people, vehicles, animals, and drones; the target behavior includes features such as being stationary, loitering, climbing, and throwing; the target distance is the vertical distance between the target and the virtual boundary in the panoramic image, with closer distances indicating higher intrusion target risk; the target dwell time is the time the intrusion target spends within the boundary defense zone, typically in seconds, with longer dwell times indicating higher intrusion target risk.

[0161] For example, a deep learning-based multimodal target detection algorithm includes: first, generating an event code based on the UUID standard and attaching spatiotemporal coordinate labels; then, extracting a multispectral fusion feature vector, including RGB spectrum distribution, infrared radiation intensity, and morphological contour parameters; simultaneously activating a time-series analysis engine, combining it with the device's historical image database to reconstruct motion trajectories and match behavior patterns, and establishing a three-dimensional spatial motion state model of the target; finally, outputting structured feature dataset parameters, including target type, target behavior, target distance, and target dwell time.

[0162] Establishing a three-dimensional spatial motion situation model of the target includes:

[0163]

[0164] In the formula, X, T, K, t represent the target type motion value, target behavior motion trend motion value, target distance, and target dwell time, respectively; G represents the target motion matching parameter; J represents the multispectral fusion feature sparsity; z represents the vertical target motion region; F represents the multispectral fusion feature vector; x represents the horizontal target motion region; y represents the vertical target motion region; and H represents the target motion region. z I is the vertical target tendency motion coefficient, and I is the radius of the target's trajectory.

[0165] S3 receives and updates the transmitted infrared and visible light panoramic images in real time as the perimeter panoramic base map. It uses a fuzzy logic algorithm to classify the intrusion target warning information and determine the priority of the intrusion target warning level. It then pushes the intrusion target warning level after priority determination via the HTTP protocol.

[0166] The intelligent alarm classification and processing center subscribes to MQTT events and receives and updates infrared and visible light panoramic images sent by the dual-band panoramic monitoring subsystem in real time as the perimeter panoramic base map. When the AI ​​intelligent analysis center identifies an intrusion target, the intelligent alarm classification and processing center receives the intrusion target warning information sent by the AI ​​intelligent analysis center, classifies the intrusion target warning level through fuzzy logic algorithm, judges the priority of the intrusion target warning level, and pushes the intrusion target warning level after priority judgment to the linkage control center through HTTP protocol.

[0167] For example, in the priority judgment of intrusion target warning level, the warning level of the intrusion target with the highest danger can be obtained.

[0168] For example, the fuzzy logic algorithm is an algorithm based on fuzzy set theory and fuzzy inference. It is a mathematical tool for handling uncertainty and fuzzy information. It transforms precise input quantities into fuzzy quantities through membership functions, performs logical inference based on a fuzzy rule base, and finally outputs a decision value through defuzzification. Its core process includes: fuzzification (mapping precise input quantities to fuzzy sets), establishing a fuzzy rule base (defining the logical relationship between input and output variables), fuzzy inference (performing fuzzy logic operations, such as AND / OR), and defuzzification (converting fuzzy outputs into precise values). The application process of the fuzzy logic algorithm in the airport perimeter system for early warning level classification is as follows:

[0169] S301, Design of fuzzification of intrusion target feature variables;

[0170] The expression for the membership function of the target type is:

[0171]

[0172] In the formula, μ type ω represents the membership degree of the target type, ranging from [0,1]; type represents the target type risk weight, ranging from [0,1]; reliability represents the confidence level of the target behavior.

[0173] The target type hazard level can be dynamically set and stored in the system database with reference to actual data and expert opinions. For example, the hazard level for a person is 1.0 and the hazard level for a vehicle is 0.8. The reliability is the target type confidence level, ranging from [0,100]. It is obtained after the target type characteristics are analyzed by the AI ​​intelligent analysis center.

[0174] The membership function expression for the target behavior is:

[0175]

[0176] In the formula, μ behavior ω represents the membership degree of the target behavior, ranging from (0,1]; behavior The target behavior risk weight is defined in the range [0,1].

[0177] The risk level of the target behavior can be dynamically set and stored in the system database with reference to actual data and expert opinions. For example, the risk level of climbing is 1.0 and the risk level of throwing is 0.7. The reliability is the confidence level of the target behavior, ranging from [0,100]. It is obtained after analyzing the characteristics of the target behavior through the AI ​​intelligent analysis center.

[0178] The target distance membership function expression is:

[0179]

[0180] where μ distance is the membership degree of the target distance, with a range of [0, 1]; d is the vertical distance between the target and the perimeter wall, m is the center point, m = 5; σ is the standard deviation, σ = 5;

[0181] d is the vertical distance between the target and the perimeter wall. When 0 ≤ d ≤ 5 m, the membership degree of the target distance is 1; when d > 20 m, the membership degree of the target distance is 0; when 5 < d ≤ 20 m, the membership degree of the target distance decreases non-linearly with the increase of the distance. According to the actual situation of the airport perimeter, the Gaussian function is used as the membership function of the warning area.

[0182] The expression of the membership function of the target residence time is:

[0183]

[0184] where μ time is the membership degree of the target residence time, with a range of [0, 1]; k is the slope coefficient, k = 0.3; T mid is the midpoint time, T mid = 10; T crit is the critical time, T crit = 20.

[0185] It increases non-linearly with the increase of the residence time of the intrusion target in the perimeter defense area. When it exceeds T crit time, μ time is set to 1. The membership function of the target residence time uses the Sigmoid function to calculate the membership degree of the intrusion target. Among them, the parameter k is the slope coefficient, which can control the steepness of the S function, that is, the severity of the change of μ time with time; the parameter T mid is the midpoint time, that is, the center symmetry point of the S function. At this time position, μ time is 0.5; T crit is the critical time, that is, the demarcation point where the S function transitions to a constant value. The function parameters can be flexibly and dynamically configured according to the actual situation of the airport perimeter and saved in the database. Generally, the function parameters are set as: k = 0.3, T mid = 10, T crit = 20.

[0186] S302, Design of the intrusion alarm fuzzy rule base;

[0187] A fuzzy rule base is a formalized representation of domain knowledge and experience. Its function is to encode and store the knowledge and experience accumulated by human experts in a specific domain in the form of "if...then..." rules. The fuzzy rule base is designed as a core component of fuzzy logic algorithms, providing the foundation and basis for fuzzy logic reasoning by storing and managing the rule information required for fuzzy inference. The design of the fuzzy rule base should satisfy the following rules:

[0188] (1) The fuzzy rule base should cover all input combination states to ensure there are no decision blind spots;

[0189] (2) The four characteristic variables of airport perimeter intrusion targets—target type, target behavior, target distance, and target dwell time—are divided into full spatial categories.

[0190] (3) Natural language description rules are adopted, namely the IF-THEN structure, and the expression form is:

[0191] IF(μ type ∈L type AND (μ) behavior ∈L behavior AND (μ) distance ∈L distance AND (μ) distance

[0192] ∈L time THEN Warning Level = C m

[0193] Where L is the fuzzy level of the feature variable, and C m The rule outputs the ultra-high / high / medium / low warning levels; L type For the fuzzy level of the feature variable of the target type, L behavior For the fuzzy level of the characteristic variable of the target behavior, L distance For the fuzzy level of the feature variable of target distance, L time Fuzzy ranking of the characteristic variable of target residence time;

[0194] (4) The fuzzy rule strength calculation should follow the AND calculation method to calculate the rule strength information that the intrusion target characteristic variables conform to. The expression is:

[0195] μ rulei =min(μ type ,μ behavior ,μ distance ,μ time )×ω i

[0196] In the formula, μ rulei To adapt the activation strength of the i-th rule to the intrusion target, ωi Let be the rule weight of the i-th rule.

[0197] The fuzzy rule base, once designed, is saved to a database and can be optimized and upgraded based on the actual operation of the airport perimeter to dynamically meet the security needs of the airport perimeter. The fuzzy rule strength calculation formula is combined with the actual situation of the perimeter to adapt to its practical use.

[0198] S303, Fuzzy Reasoning and Defuzzification of Intrusion Alarm Features;

[0199] Fuzzy inference of intrusion alarm features is based on fuzzy feature variables and fuzzy rule base, and takes the feature variables of the intrusion target as input. It performs inference by following the steps of feature variable fuzzification, rule strength evaluation, rule consequent aggregation and defuzzification, and finally obtains the warning level of the intrusion target.

[0200] For example, the defuzzification is a key step in fuzzy logic reasoning. Its function is to aggregate the fuzzy results of multiple rules based on the fuzzified feature variables (such as the membership values ​​of target type, behavior, distance, and dwell time) and the fuzzy rule base through logical operations, and finally output a single, definite warning level so that the system can execute the corresponding linkage strategy.

[0201] Among these steps, feature variable fuzzification is a crucial preliminary step in fuzzy logic reasoning. Its core is to transform the precise feature data of the intrusion target (such as distance 5.2 meters and dwell time 15 seconds) into membership values ​​in a fuzzy set (such as a membership degree of 0.8 for "close distance"), providing a foundation for subsequent rule-based reasoning. The steps include: determining the input feature variables, designing the membership function, and fuzzy level classification.

[0202] For example, the rule strength evaluation is to perform logical operations on each rule antecedent (precondition) according to the fuzzy rule strength calculation formula based on a predefined fuzzy rule base (IF-THEN rules) to obtain the activation strength of the rule consequent (conclusion);

[0203] The rule consequent aggregation merges the fuzzy sets of all activated rule consequents into a comprehensive output fuzzy set, using the maximum value method (MAX) according to arg max(μ). rulei The results of multiple rules are superimposed and merged in a manner that allows for the aggregation and merging of μ. rule =max(μ rule1 ,μ rule2 ,μ rule3 …μ rulen The alarm rule strength value is obtained, and then dynamically divided according to the activation strength threshold maintained by the system. Common dynamic optimization methods include manual configuration, data-driven optimization, and real-time adaptation.

[0204] Manual configuration includes: manually adjusting the threshold range through the management interface (e.g., narrowing the "high-level" threshold range in sensitive airport areas to improve response sensitivity);

[0205] Data-driven optimization includes: automatically fitting optimal thresholds based on historical intrusion event data through machine learning (such as clustering algorithms) to adapt to the environmental differences of different airports;

[0206] Real-time adaptation includes dynamically scaling the threshold range by combining real-time weather, passenger flow, and other data (e.g., increasing false alarm tolerance and temporarily expanding the "low-level" range during heavy rain). The activation intensity range of the warning level generally includes four levels, such as: ultra-high [0.8,1], high [0.6,0.8], medium [0.3,0.6], and low [0.0,0.3].

[0207] S304, the intelligent alarm classification processing center saves the intrusion target warning level obtained by fuzzy algorithm reasoning to the database, and pushes the intrusion target coordinates and intrusion target warning level to the linkage control center via HTTP protocol.

[0208] S4, based on the priority judgment of the push notification and the warning level of the intrusion target, uses the OPENCV algorithm library to mark the trajectory of multiple nearest coordinate points of the intrusion target in the perimeter panoramic base map, displays the current movement trajectory of the intrusion target, and judges the intrusion target's intent.

[0209] For example, the coordinates of the intrusion target are the pixel positions of the intrusion target in the panoramic image. The position of the intrusion target may change every time the scanning device rotates, thereby causing the intrusion target warning level to change.

[0210] For example, the OPENCV algorithm library is an open-source algorithm library for computer vision, which is efficient and cross-platform. It can perform various processing and operations on images. Compared with the traditional airport perimeter system, which can only mark the real-time location of the intrusion target and lacks historical trajectory association and motion trend analysis, this invention uses multi-frame coordinate association, visualization enhancement, real-time optimization and other methods to visualize the motion trajectory of the intrusion target in real time, and assists airport perimeter staff in judging the intention of the intrusion target.

[0211] The trajectory annotation process includes: 1) Obtaining the six most recent valid real-time coordinates (PositionX, PositionY) of the intrusion target in chronological order to avoid interference from outdated data; 2) Reading a dual-band panoramic image (visible light / infrared) using cv2.imread() as the base image for trajectory annotation; 3) Drawing line segments point by point on the six cached coordinates in chronological order using the cv2.line() function; 4) Setting the line color according to the target's warning level (e.g., red for ultra-high level, yellow for high level) to enhance visual differentiation; 5) Triggering the trajectory annotation process based on the latest panoramic image generated by the perimeter scanning device to automatically update the trajectory. Furthermore, during trajectory annotation, OpenCV's cv2.GaussianBlur() is used to smooth the coordinates, filtering out noise points caused by device jitter or temporary target occlusion. This completes real-time trajectory tracking of the intrusion target under the panoramic image.

[0212] For example, the joint control center performs trajectory tracking and video tracking of the intrusion target based on the target's coordinates and warning level. The trajectory tracking algorithm (connecting the six nearest coordinates of the intrusion target using OpenCV to form the target's trajectory) calculates and connects the real-time coordinates of the intrusion target, displaying the trajectory in real-time on the front-end monitoring system to help monitoring personnel determine the intrusion target's intent. The video tracking uses a coordinate mapping algorithm to map the real-time intrusion target's coordinates to PTZ parameters on a PTZ camera, and calls the PTZ camera SDK in real-time to adjust the camera's angle and zoom level based on the target's coordinates, thus enabling real-time video tracking of the intrusion target.

[0213] For example, the linkage control center can automatically identify and lock onto the most threatening target based on the intrusion target warning level. By mapping the intrusion target's coordinates on the perimeter panoramic base map to the PTZ camera's horizontal rotation angle (P), pitch angle (T), and zoom magnification (Z) in real time, precise tracking of the intrusion target is achieved. Simultaneously, the linkage control center retrieves PTZ camera information bound to the current perimeter scanning equipment from the database and constructs a three-dimensional spatial coordinate transformation model based on the mapping relationship between the panoramic map coordinate system and the PTZ camera's physical coordinate system. This model automatically controls the PTZ camera to turn towards the intrusion target for video tracking and confirmation.

[0214] The PTZ (Pan-Tilt-Zoom) type PTZ camera is widely used in the field of security monitoring and has horizontal rotation (Pan), vertical tilt (Tilt), and lens zoom (Zoom) functions.

[0215] The initial horizontal angle of the PTZ camera is 0°, consistent with the initial orientation setting of the peripheral scanning equipment. The linkage control center calculates the PTZ parameters of the PTZ camera using the coordinates of the intruding target (PositionX, PositionY), the vertical field of view (ForVer), and the pitch angle (DangleVer) of the peripheral scanning equipment. The specific mapping formula is as follows:

[0216] The horizontal angle (P) is calculated using a polar coordinate mapping algorithm, converting the target pixel coordinates (PositionX / ImgWidth) into the PTZ camera's horizontal rotation angle.

[0217]

[0218] In the formula, ImgWidth and ImgHeight are the pixel width and height of the panoramic image, respectively, and θ offset The initial angle calibration offset between the peripheral scanning equipment and the PTZ camera is pre-measured and stored through the equipment calibration program. This offset is applied when the peripheral scanning equipment and the PTZ camera are at the same angle θ. offset =0;

[0219] Vertical tilt (Tilt) is calculated using a trigonometric algorithm, combining the PTZ camera's vertical field of view (ForVer) and tilt angle (DangleVer).

[0220]

[0221] In the formula, the negative sign indicates the engineering convention that the downward direction of the PTZ camera is the positive direction;

[0222] The formula essentially converts the vertical offset of a pixel into a pitch angle in physical space.

[0223] The zoom factor Z is:

[0224]

[0225] In the formula, Z min Z max These are the minimum and maximum zoom levels preset for the PTZ camera, respectively.

[0226] This formula ensures that the zoom level automatically increases when the target is below the panoramic view (at close range), enabling special monitoring.

[0227] In summary, the linkage control center sends PTZ positioning commands to the PTZ camera based on real-time calculated PTZ parameters. The PTZ camera's actuator employs a trajectory planning algorithm based on a trapezoidal velocity curve to achieve smooth control of horizontal (Pan) and vertical (Tilt) movement. Upon receiving the command, it performs smooth horizontal and vertical movement according to the P and T parameters. After reaching the target position, it performs zoom processing based on the Z parameters and automatically adjusts the focus to obtain a clear image of the intrusion target.

[0228] As can be seen from the above embodiments, the present invention can reduce the manpower cost of airport security: by using intelligent hierarchical early warning and linkage tracking, the frequency of manual patrols can be reduced and the emergency response time can be shortened.

[0229] Market potential for equipment upgrades: Traditional airport perimeter systems (such as fiber optic and infrared beam systems) have technical bottlenecks, and this invention can serve as a next-generation intelligent transformation solution. It is compatible with over 200 airports nationwide.

[0230] Existing airport perimeter security systems primarily rely on single technologies (such as electronic fences and video surveillance). While some systems incorporate AI detection, tiered early warning systems depend on fixed thresholds (e.g., "high alarm triggered within 5 meters of the perimeter") and cannot dynamically integrate multi-dimensional features (such as behavioral patterns and dwell time). This invention is the first to integrate fuzzy logic reasoning into the field of perimeter security, enabling dynamic quantitative assessment of intrusion risks and filling the technological gap in "multi-feature fusion and adaptive tiering."

[0231] Existing systems largely rely on staff to determine the danger level of intrusion targets, resulting in serious false alarms and missed alarms, leaving staff overwhelmed. This invention uses fuzzy logic to combine and reason about target type, behavior, distance, and dwell time to automatically determine high, medium, and low risk, reducing manpower costs and avoiding resource waste. Moreover, the intelligent perimeter warning technology based on fuzzy logic reasoning enables accurate response to intrusion targets, solving the industry pain point that "static rules cannot adapt to dynamic threats."

[0232] Traditional approaches assume that "high-precision sensors (such as LiDAR) and fixed-threshold algorithms" are sufficient, but they overlook the diversity of threats and the complexity of environments in real-world scenarios. This invention breaks this bias, demonstrating that an intelligent perimeter early warning scheme based on fuzzy logic reasoning is superior in both adaptability and accuracy.

[0233] Example 2, as Figure 1 The airport perimeter hierarchical early warning and linkage system based on fuzzy logic reasoning provided in this embodiment of the invention mainly consists of four core modules: a dual-band panoramic monitoring subsystem, an AI intelligent analysis center, an intelligent alarm hierarchical processing center, and a linkage control center.

[0234] Another objective of this invention is to provide a hierarchical early warning and linkage system for airport perimeter based on fuzzy logic reasoning, comprising:

[0235] The dual-band panoramic monitoring subsystem is used to scan the airport perimeter to generate infrared and visible light panoramic images, and to send the generated infrared and visible light panoramic image information to the outside world via the MQTT network protocol.

[0236] The AI ​​intelligent analysis center receives infrared and visible light panoramic image data and uses a deep learning-based multimodal target detection algorithm to detect and identify intrusion targets in the perimeter defense zone of the infrared and visible light panoramic images. It then pushes out early warning information for the detected and identified intrusion targets, which includes the coordinates of the intrusion target and its intrusion characteristics.

[0237] The intelligent alarm classification and processing center is used to receive and update infrared and visible light panoramic images in real time as the perimeter panoramic base map. It classifies the intrusion target warning information through fuzzy logic algorithm, judges the priority of the intrusion target warning level, and pushes the intrusion target warning level after priority judgment through HTTP protocol.

[0238] The linkage control center is used to determine the priority of the push notifications and the warning level of the intrusion target. It uses the OPENCV algorithm library to mark the trajectory of multiple nearest coordinate points of the intrusion target on the perimeter panoramic base map, display the current movement trajectory of the intrusion target, and determine the intrusion target's intent.

[0239] For example, the dual-band panoramic monitoring subsystem includes multiple sets of perimeter scanning devices, and stores the parameter information of each perimeter scanning device in a database. The perimeter scanning device is a monitoring device capable of panoramic scanning, possessing a 360° horizontal viewing angle, an 8.7° vertical viewing angle, a pitch angle range of -20° to +80°, and a scanning speed of 144° / s (i.e., completing one rotation in 2.5 seconds). Its image detector resolution is 26288x640 in infrared mode and 22692x512 in visible light mode. Each perimeter scanning device is installed at 200-meter intervals and fixed to the crossarm of a pole along with the PTZ camera. The initial horizontal turning angle of the perimeter scanning device is consistent with that of the PTZ camera, and the pole is installed on the inner side near the perimeter wall. The parameter information of the perimeter scanning device includes key data such as device number, IP address, port number, longitude, latitude, visible light panoramic image width, visible light panoramic image height, infrared panoramic image width, infrared panoramic image height, vertical field of view, and pitch angle.

[0240] like Figure 2 The dual-band panoramic monitoring subsystem establishes a network connection with the perimeter security system service via a network switch. After the weekly scanning device completes a week's scan, the system sends a panoramic image of the device to the AI ​​intelligent analysis center and the intelligent alarm classification and processing center via the MQTT protocol, and saves the message to the perimeter security system database. The MQTT message's topic is / img / full / realtime, and the panoramic image message is transmitted in JSON format.

[0241] For example, after completing a scan on the weekly scanning device numbered 001, the message format is as follows:

[0242] {deviceId:001,imgs:[{channelId:0,width:26288,height:640,forVer:8.7,dangleVer:-20.0,img:byte[]},{c channelId:1,width:22692,height:512,forVer:8.7,dangleVer:-20.0,img:byte[]}],time:2024-03-1510:00:00}

[0243] [deviceId is the device ID; imgs is an array of panoramic images from the device, containing infrared and visible light panoramic images, where channelId is the channel number, 0 is the infrared panoramic image channel, 1 is the visible light panoramic image channel, width is the panoramic image width, height is the panoramic image height, forVer is the vertical field of view of the device, dangleVer is the pitch angle of the device, img is the panoramic image information, stored as a byte array; time is the message upload time].

[0244] The AI ​​intelligent analysis center subscribes to the MQTT topic ( / img / full / realtime) to obtain real-time panoramic image data streams of devices pushed by the dual-band panoramic monitoring subsystem. The system uses a multi-threaded mechanism to simultaneously analyze visible light and infrared dual-mode panoramic images, and sets up virtual perimeter walls based on the panoramic image information. A core perimeter defense zone is set 5 meters inside and outside the perimeter wall, and a perimeter early warning defense zone is set 20 meters inside and outside the perimeter wall.

[0245] Meanwhile, the AI ​​intelligent analysis center uses a deep learning-based multimodal target detection algorithm to detect and identify intrusion targets in real-time panoramic images, and eliminates static environmental interference through an adaptive background modeling algorithm. When a valid intrusion target is triggered, the system executes a multi-dimensional analysis process: first, it generates an event code based on the UUID standard and attaches a spatiotemporal coordinate label; then, it extracts a multispectral fusion feature vector (including RGB spectrum distribution, infrared radiation intensity, and morphological contour parameters); simultaneously, it activates the time-series analysis engine, combines the device's historical image library to reconstruct motion trajectories and match behavioral patterns, and establishes a three-dimensional spatial motion situation model of the target. The final output structured feature dataset covers the following parameters: target type (classification code: 00-Other / 01-Personnel / 02-Vehicle / 03-Animal / 04-Drone), target behavior (state code: 00-Still / 01-Wandering / 02-Throwing / 03-Climbing), target distance (the shortest spatial Euclidean distance based on the boundary wall reference plane, in meters), and target dwell time (the continuous dwell time of a target with the same number within the defense zone, in seconds).

[0246] Furthermore, after the AI ​​intelligent analysis center completes the extraction of intrusion target feature information, it calls the WebAPI interface ( / api / analysis) of the intelligent alarm hierarchical processing center via HTTP protocol to upload the intrusion target feature information to the intelligent alarm hierarchical processing center. The intrusion target feature information data is transmitted in JSON string format.

[0247] For example, based on the intrusion target feature information extracted from the panoramic image of the No. 001 weekly scanning device, the following message is sent:

[0248] {alarmId:00001,timestamp:1742108033,targets:[{channelId:1,targetId:001,origi nalPositionX:12634,originalPositionY:68,pixelHeight:26,pixelWidth:8,type:{value:01,reliability:80},behavior:{value:01,reliability:90},distance:10,time:11},{channelId:1,targ etId:002,originalPositionX:4355,originalPositionY:398,pixelHeight:142,pixelWidth:54,type:{value:01,reliability:95},behavior:{value:02,reliability:85},distance:5,time:19}]};

[0249] Wherein, alarmId is the alarm number; timestamp is the current timestamp; targets is an array of intrusion targets, which can contain information on one or more intrusion targets. Each intrusion target contains the following: channelId is the channel number, 0 for the infrared panoramic image channel and 1 for the visible light panoramic image channel; targetId is the intrusion target number; originalPositionX is the intrusion target's x-coordinate pixel value; originalPositionY is the intrusion target's y-coordinate pixel value; pixelHeight is the intrusion target's height pixel value; pixelWidth is the intrusion target's width pixel value; type is the intrusion target type, including content and confidence level; behavior is the intrusion target behavior, including content and confidence level; distance is the distance between the intrusion target and the nearest perimeter wall, in meters (m); time is the intrusion target's dwell time in the perimeter defense zone, in seconds (s).

[0250] For example, the intelligent alarm grading and processing center receives device panoramic image messages uploaded by the dual-band panoramic monitoring subsystem in real time by subscribing to the MQTT topic ( / img / full / realtime), and stores the visible light and infrared panoramic images separately in a distributed file server. Perimeter staff can select specific perimeter scanning devices through a visual front-end interface to view the updated panoramic image information in real time.

[0251] After receiving the intrusion target characteristic information from the AI ​​intelligent analysis center, the intelligent alarm classification and processing center uses fuzzy logic algorithms to perform logical reasoning on the target characteristics, ultimately determining the warning level of the intrusion target. The logical reasoning process is as follows: Figure 3 As shown, the specific steps are as follows:

[0252] First, the input variables are defined in a fuzzy manner, and the characteristics of the intrusion target are fuzzified. According to a preferred embodiment of this example, the fuzziness level of the input variables is set through the perimeter system. For example, the fuzziness level of the target type (high: ≥0.8, medium: 0.5-0.8, low <0.5) corresponds to the following types: human (1.0) = high, drone (0.8) = high, vehicle (0.6) = medium, animal (0.3) = low; the fuzziness level of the target behavior (high: ≥0.8, medium: 0.5-0.8, low <0.5) corresponds to the following types: human (1.0) = high, drone (0.8) = high, vehicle (0.6) = medium, animal (0.3) = low. 0.5) Corresponding behaviors: Climbing (1.0) = High, Throwing (0.9) = High, Loitering (0.7) = Medium, Still (0.4) = Low; Target distance fuzziness level (near: ≥0.75, medium: 0.3-0.75, far: <0.3), parameters m=5, σ=5 in the distance formula; Target dwell time fuzziness level (long: ≥0.7, medium: 0.4-0.7, short: <0.4), parameters k=0.3, Tmid=10, Tcrit=20 in the dwell time formula. After the fuzziness level of the input variables is defined, the membership degree of each intrusion target feature can be calculated according to the membership degree calculation formula, for example, calculating the membership degree of the intrusion target feature information extracted from the panoramic image of the 001 perimeter scanning device. The membership degrees of target 001 are as follows: target type membership degree μ_type=ω_type×(reliability / 100)=1.0×(80 / 100)=0.80; target behavior membership degree μ_behavior=ω_behavior×(reliability / 100)=0.7×(90 / 100)=0.63; the target distance is 10 meters, so the target distance membership degree μ_distance=e^[-(dm)^2 / 2σ^2]=e^[-(10-5)^2 / 2×5^2]≈0.60; the target dwell time is 11 seconds, so the target dwell time membership degree μ_time=1 / (1+exp[-k(t-Tmid)])=1 / (1+exp[-0.3(11-10)])≈0.57. Similarly, the membership degrees of target 002 are calculated as follows: target type membership degree μ_type = 1.0 × (95 / 100) = 0.95; target behavior membership degree μ_behavior = 0.9 × (85 / 100) = 0.85; target distance membership degree μ_distance = 1.0 (5 meters is within the perimeter alarm zone, according to the target distance membership degree formula, when it is less than or equal to 5 meters, the membership degree = 1.0); target dwell time membership degree μ_time = 1 / (1 + exp[-0.3(19-10)]) ≈ 0.94. Therefore, the final set of membership degrees of target 001 is obtained (0.80 = high, 0.63 = medium, 0.60 = medium, 0.57 = medium), and the set of membership degrees of target 002 is obtained (0.95 = high, 0.85 = high, 1.0 = near, 0.94 = long).

[0253] Then, based on knowledge and expert experience in the perimeter security field, an intrusion alarm fuzzy rule base is constructed. The intrusion alarm fuzzy rule base should cover all possible input combinations and perform full-space partitioning of intrusion target feature variables, wherein the rule base design adopts an IF-THEN structure description. According to another preferred embodiment of this example, a partial rule base matrix is ​​as follows:

[0254] R0001: IF μ_type = High AND μ_behavior = High AND μ_distance = Near THEN Output level = Very High, weight ω1 = 1.0;

[0255] R0002: IF μ_type = High AND μ_behavior = High AND μ_distance = Medium AND μ_time = Long THEN Output level = High, weight ω2 = 0.9;

[0256] R0003: IF μ_type = High AND μ_distance = Near THEN Output level = High, weight ω3 = 0.8;

[0257] R0004: IF μ_type = High AND μ_behavior = High AND μ_distance = Medium AND μ_time = Medium THEN Output level = High, weight ω4 = 0.8;

[0258] R0005: IF μ_type = High AND (μ_behavior = Medium OR μ_time = Medium) AND μ_distance = Medium THEN Output level = Medium, weight ω5 = 0.6;

[0259] R0006: IF μ_type = High AND μ_behavior = Medium AND μ_distance = Far AND μ_time = Short THEN Output level = Low, weight ω6 = 0.3;

[0260] Furthermore, based on the defined fuzzy rule base and the fuzzy rule activation intensity calculation formula, rule evaluation is performed on the feature membership sets of target 001 and target 002 respectively. Target 001 fits rule R0005, therefore the activation intensity of target 001 is μ_rule=max(min(μ_type,μ_behavior,μ_distance),min(μ_type,μ_time,

[0261] μ_distance)) × ω5 = max (min (0.80, 0.63, 0.60), min (0.80, 0.57, 0.60)) × 0.6 = 0.36;

[0262] Target 002 is adapted to both rules R0001 and R0003. Therefore, the activation intensity of target 002 is μ_rule=max(μ_rule1,μ_rule3)=max(min(0.95,0.85,1.0,0.94)×1.0,min(0.95,1.0)×0.8)=0.85.

[0263] Finally, according to the warning level intensity classification rules maintained by the system, the final aggregated activation intensity value of the intrusion target is deblurred. According to another preferred embodiment of this example, the warning level intensity classification is ultra-high [0.8,1], high [0.6,0.8], medium [0.3,0.6], and low [0.0,0.3]. Therefore, the warning level of target 001 is 0.36 = medium, and the warning level of target 002 is 0.85 = ultra-high.

[0264] For example, after the intelligent alarm classification processing center performs fuzzy logic reasoning on the intrusion target alarm information and finally derives the final warning level, the linkage control center will link the corresponding PTZ camera to conduct video monitoring and tracking of the intrusion target with the highest warning level, and at the same time automatically alert staff according to the perimeter security handling procedure. According to the intrusion target warning level calculation in this embodiment, target 002 is the target with the highest warning level. Simultaneously, based on the target 002 coordinate information (4355, 20) and the PTZ parameter mapping formula of the PTZ camera, P = (PositionX / ImgWidth) × 360 = (4355 / 22692) × 360 ≈ 69.1; T = -(DangleVer + arctan([2 × |ImgHeight / 2 - PositionY| × tan(ForVer / 2)] / ImgHeight)) ≈ 22; Z = ZoomMin + (ZoomMax - ZoomMin) × [(ImgHeight - PositionY) / ImgHeight] 2 =2 + (20 - 2) × [(512 - 398) / 512]² ≈ 3. Therefore, the PTZ parameters of the PTZ camera are (69.1, 22, 3). The linkage control center sends the PTZ parameter values ​​to the PTZ camera. The PTZ camera moves according to the parameters in this example, specifically with a horizontal movement angle of 69.1°, a vertical angle of 22° (the downward direction of the PTZ camera is positive), and a magnification of 3x.

[0265] Another example is that staff can use the front-end alarm monitoring interface to intuitively view real-time visible light and infrared panoramic images from various perimeter scanning devices, while simultaneously obtaining various intrusion target warning levels and characteristic information reported by the intelligent alarm classification and processing center. Furthermore, the alarm monitoring interface can display the movement trajectory of intrusion targets in real-time within the panoramic image and automatically provide a real-time video monitoring page for the intrusion target with the highest warning level. Of course, staff can also manually operate the PTZ camera to observe and monitor other intrusion targets as needed. Through these functions, staff can accurately determine the intentions of intrusion targets, thereby promptly carrying out emergency response and security control work for intrusion alarms, ensuring the safety and stability of the relevant areas.

[0266] To further illustrate the effects of the embodiments of the present invention, the following experiments were conducted.

[0267] To verify the analytical processing effect of this invention, a simulated real airport perimeter intrusion scenario was used. The method of this invention was applied to verify its ability to identify targets of different risk levels, the accuracy of warning level judgment, and the effectiveness of linked PTZ camera video tracking and trajectory annotation. A simulation program was implemented and a simulation experiment was conducted. First, a simulation experiment was conducted using multiple historical infrared and visible light panoramic images of airport perimeters. Multiple historical infrared and visible light panoramic images of airport perimeters were collected, covering different time periods, weather conditions, and lighting conditions to ensure the diversity of image data. Detailed annotations were made of target information in the images, including target type (person, vehicle, animal, drone), target location (accurate to meters, range [0m, 50m]), target behavior (stationary, loitering, climbing, throwing), and target dwell time (accurate to seconds, range [0s, 30s]). Furthermore, parameters and rules are defined for fuzzy logic reasoning. For example, the fuzzy level for target type is set as (high: ≥0.8, medium: 0.5-0.8, low <0.5), the fuzzy level for target behavior is set as (high: ≥0.8, medium: 0.5-0.8, low <0.5), the fuzzy level for target distance is set as (near: ≥0.75, medium: 0.3-0.75, far: <0.3), the fuzzy level for target dwell time is set as (long: ≥0.7, medium: 0.4-0.7, short: <0.4), the number of fuzzy rules is set to 64, and the alarm intensity is set as ultra-high [0.8,1], high [0.6,0.8], medium [0.3,0.6], and low [0.0,0.3]. Based on the above fuzzy levels, combined with practical experience and domain knowledge of airport perimeter security, 64 fuzzy rules are constructed. Simultaneously, alarm intensity levels are set as ultra-high [0.8,1], high [0.6,0.8], medium [0.3,0.6], and low [0.0,0.3]. A simulation program is written to generate 100 sets of simulated data using collected historical image data, according to the target type, location, behavior, and dwell time requirements. The system then simulates and pushes perimeter infrared and visible light panoramic images. This invention achieves panoramic image processing, target feature extraction, fuzzy logic reasoning, warning level judgment, PTZ camera linkage control, and trajectory annotation. During processing, the membership function parameters are continuously optimized and adjusted to achieve the best target recognition effect.

[0268] In this simulation, 11 targets were identified as extremely high-risk, 24 targets as high-risk, 36 targets as medium-risk, 15 targets as low-risk, and 14 targets posed no danger, thus not triggering any perimeter alarms. By comparing the alarm levels identified in the simulation with the preset actual risk levels, the accuracy of the alarm level judgment was high, the average deviation rate of the linked PTZ camera video tracking was low, and the trajectory marking and tracking effect was good. Meeting the actual needs of airport security, this invention more effectively assists perimeter monitoring personnel in quickly identifying and monitoring high-risk targets compared to traditional perimeter systems.

[0269] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An airport boundary grading early warning linkage method based on fuzzy logic reasoning, characterized in that, The method comprises the following steps: S1, panoramic scanning of the airport perimeter to generate infrared and visible light panoramic images, and sending the generated infrared and visible light panoramic image information to the outside through an MQTT network protocol; S2, receiving infrared and visible light panoramic image data, and detecting and identifying intrusion targets in the perimeter defense area in the infrared and visible light panoramic images through a deep learning-based multi-modal target detection algorithm; the detected and identified intrusion target warning information is pushed, and the intrusion target warning information includes an intrusion target coordinate point and an intrusion target intrusion feature information; S3, receiving and updating the sent infrared and visible light panoramic images as a perimeter panoramic base map in real time, dividing the sent intrusion target warning information into an intrusion target warning level through a fuzzy logic algorithm, judging the priority of the intrusion target warning level, and pushing the intrusion target warning level after the priority judgment through an HTTP protocol; S4, using an OPENCV algorithm library to mark the trajectories of multiple coordinate points closest to the intrusion target in the perimeter panoramic base map for the pushed intrusion target warning level after the priority judgment, displaying the motion trajectory of the current intrusion target, and judging the intention of the intrusion target; In step S1, the infrared and visible light panoramic images are generated by a dual-band panoramic monitoring subsystem scanning the airport perimeter; the dual-band panoramic monitoring subsystem comprises a plurality of panoramic scanning perimeter monitoring devices, each of which is installed at an interval and is installed on a horizontal arm of a rod together with a dome camera, and the rod is installed inside the airport perimeter; each device generates a 360-degree panoramic image at a fixed time interval through continuous horizontal 360-degree rotation; Step one, establish a weakening mechanism model of the interference light disturbance grayscale image, and determine the optimization framework; Step two, construct a data item for modeling the interference light disturbance; Step three, select an image prior item, and construct a non-optimization feedback model of the interference light disturbance grayscale image in combination with the data item; Step four, obtain a clear panoramic image by using an iterative reweighted second-order iterative optimization method.

2. The airport boundary grading early warning linkage method based on fuzzy logic reasoning according to claim 1, characterized in that, In step one, the weakening mechanism model of the interference light disturbance grayscale image is established, and the optimization framework is determined, including: (1) a nonlinear weakening model of the interference light disturbance grayscale image is established, and the expression is: ； ； wherein, is the vector of the blurred image not contaminated by the gray scale pollution, is the known blur kernel, is the vector of the original sharp image, is the vector of the final attenuated image, is the process of the image being contaminated by the light disturbance. (2) according to the weakening model, the data fidelity and the image prior are combined into a unified image restoration framework based on the maximum a posteriori probability framework, and the non-optimization feedback problem corresponds to an optimization problem, and the expression is: ； In the formula, to unify the image vector in the image restoration framework, is the posterior probability; According to the Bayes formula, we have: ； wherein is the likelihood probability, is the prior probability; Joint the above two formulas, and equivalent transformation is: ； In the formula, is the equivalent transformed image vector; The following optimization framework is obtained: ； wherein is a data item to be constructed, is an image prior to be designed; In step two, the data item for modeling the interference light disturbance is constructed, specifically including: (1) the following nonlinear function is selected to remove the influence of the disturbance grayscale: ； wherein are function parameters, control the degree of non-linearity of the function at the cut-off, control the cut-off point, is a blur kernel with the effect of the dithering removed, is the Reynolds number; (2) the nonlinear function of the L1 norm is combined to construct the following data item: ； wherein is an image pixel index, is a vector of the final attenuated image, is a data item of the image vector, is a non-linear function; In step three, the non-optimization feedback model of the interference light disturbance grayscale image is constructed in combination with the data item, including: The expression of the prior item constructed according to the image gradient sparse prior is: ； In the formula, is the first image node image gradient sparse prior value, is the image gradient sparse prior non-optimization feedback, is the first image node image gradient sparse prior value; In step four, the iterative reweighted second-order iterative optimization method, including: The non-optimization feedback model is deformed into the form of the weighted second-order iterative optimization method shown in the following formula: ； wherein is a regularization term weight, is a weighted second order iterative optimization, is a first is a diagonal matrix of image gradient sparsity prior value weights in image node deformation, is a first is a blur kernel in image node deformation, is a first is a diagonal matrix of image gradient sparsity prior value weights in image node deformation, is a first is a blur kernel in image node deformation.

3. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 1, characterized in that, In step S2, infrared and visible light panoramic image data are received, including: The AI intelligent analysis center receives the infrared and visible light panoramic images sent by the dual-band panoramic monitoring subsystem by subscribing to the MQTT event, detects and identifies the intrusion targets in the perimeter defense area in the infrared and visible light panoramic images through a multi-modal target detection algorithm based on deep learning, and pushes the detected and identified intrusion target early warning information to the intelligent alarm hierarchical processing center; The content of the intrusion target detection includes target type, target behavior, target distance and target residence time detection content; wherein the target type includes the characteristics of people, vehicles, animals and unmanned aerial vehicles; the target behavior includes the characteristics of stillness, wandering, climbing and throwing; the target distance is the vertical distance feature of the target from the virtual perimeter in the panoramic image, and the closer the distance, the higher the risk of the intrusion target; the target residence time is the time feature of the intrusion target staying in the perimeter defense area; The multi-modal target detection algorithm based on deep learning includes: first, generating event encoding based on the UUID standard and attaching space-time coordinate labels; then extracting a multi-spectral fusion feature vector containing RGB spectral distribution, infrared radiation intensity and morphological profile parameters; synchronously activating a time sequence analysis engine, combining with a device historical image library to reconstruct a motion trajectory and match a behavior pattern, and establishing a target three-dimensional space motion situation model; finally, outputting a structured feature data set parameter including target type, target behavior, target distance and target residence time; Establishing a target three-dimensional space motion situation model includes: ； ； ； wherein, are target type motion values, target behavior motion trend motion values, target distance and target dwell time, respectively, is a target motion matching parameter, is a multi-spectrum fusion feature sparsity, is a vertical target motion region, is a multi-spectrum fusion feature vector, is a horizontal target motion region, is a longitudinal target motion region, is a vertical target trend motion coefficient, is a target motion trajectory radius.

4. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 1, wherein, In step S3, a fuzzy logic algorithm is used for intrusion target early warning level division and intrusion target early warning level priority judgment, including: S301, fuzzy design of intrusion target feature variables; S302, design of intrusion alarm fuzzy rule base; S303, fuzzy reasoning of intrusion alarm features and defuzzification; S304, the intelligent alarm hierarchical processing center saves the intrusion target early warning level obtained by the fuzzy algorithm reasoning to the database, and pushes the intrusion target coordinate point and the intrusion target early warning level to the linkage control center through the HTTP protocol.

5. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 4, wherein, In step S301, the fuzzy design of the intrusion target feature variables includes: The target type membership function expression is: ； wherein, is the target type membership, ranging from ; is the target type danger weight, ranging from ; is the target behavior confidence; The target behavior membership function expression is: ； wherein, is the target behavior membership, ranging from ; is the target behavior risk weight, ranging from ; The target distance membership function expression is: ； wherein is the target distance membership, ranging from ; is the vertical distance from the target distance perimeter, is the center point, ; is the standard deviation, ; The target residence time membership function expression is: ； wherein is the target residence time membership, ranging from ; is the slope coefficient, is the midpoint time, is the critical time; In step S302, the design of the intrusion alarm fuzzy rule base includes: The design of the fuzzy rule base meets the following rules: (1) The fuzzy rule base should cover all input combination states; (2) The target type, target behavior, target distance and target residence time of the airport perimeter intrusion target are divided in the whole space; (3) The natural language description rule is used, that is, the IF-THEN structure, and the expression form is: ； wherein, is a characteristic variable fuzzy level, is a rule output super high / high / medium / low warning level; is a characteristic variable fuzzy level of a target type, is a characteristic variable fuzzy level of a target behavior, is a characteristic variable fuzzy level of a target distance, is a characteristic variable fuzzy level of a target dwell time; (4) The fuzzy rule strength calculation should be calculated according to the AND calculation method, and the rule strength information that meets the intrusion target feature variables is calculated, and the expression is: ； In the formula, activating the rule of the first rule of the first rule weight of the first rule.

6. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 3, wherein, In step S303, the fuzzy reasoning of the intrusion alarm features and the defuzzification include: The fuzzy feature variable and the fuzzy rule base are used as the basis, and the intrusion target feature variable is taken as an input item, and inference is carried out according to the steps of feature variable fuzzification, rule strength evaluation, rule consequent aggregation and defuzzification, and finally the intrusion target warning level is obtained; The rule strength evaluation is logical operation according to the pre-defined fuzzy rule base, and the activation strength of the rule consequent is obtained by logical operation according to the fuzzy rule strength calculation formula for each rule antecedent; The rule consequent is aggregated, combining the fuzzy sets of all activated rule consequents into one comprehensive output fuzzy set, using the maximum method MAX to superimpose and combine the results of multiple rules in the manner The alarm rule strength value is obtained, and then dynamically divided according to the activated strength threshold maintained by the system, wherein the dynamic optimization methods commonly include manual configuration, data-driven optimization, and real-time self-adaptation.​ The artificial configuration includes manual adjustment of the threshold interval through the management interface; the data-driven optimization includes automatic fitting of the optimal threshold based on historical intrusion event data through machine learning, and adaptation to the environmental differences of different airports; Real-time self-adaptation includes dynamically scaling the threshold interval in combination with real-time weather and passenger flow data; the warning level activation strength range includes four levels, ultra-high [0.8, 1], high [0.6, 0.8), medium [0.3, 0.6), and low [0.0, 0.3].

7. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 1, wherein, In step S4, the OPENCV algorithm library is used to mark the trajectory of the intrusion target at multiple coordinate points closest to the intrusion target in the perimeter panoramic base map, to show the motion trajectory of the current intrusion target, and to judge the intention of the intrusion target, including: The linkage control center can automatically identify and lock the maximum threat target according to the intrusion target warning level; by mapping the intrusion target coordinate points of the target in the perimeter panoramic base map into the horizontal rotation angle P, the pitch angle T and the zoom factor Z of the ball machine in real time, the tracking of the intrusion target is realized; at the same time, the linkage control center obtains the ball machine information bound with the current perimeter scanning device from the database, and based on the mapping relationship between the perimeter panoramic base map coordinate system and the ball machine physical coordinate system, a three-dimensional space coordinate conversion model is constructed, the ball machine is automatically controlled to turn to the direction of the intrusion target, and video tracking confirmation is performed.

8. The airport boundary grading early warning linkage method based on fuzzy logic reasoning of claim 7, characterized in that, The linkage control center calculates the PTZ parameters of the ball machine through the coordinate points of the intrusion target , the vertical field angle ForVer and the pitch angle DangleVer of the peripheral scanning device, and the specific mapping process is as follows: The horizontal angle P adopts a polar coordinate mapping algorithm to convert the target pixel coordinate into a horizontal rotation angle of the spherical camera: ； In the formula, respectively the pixel width and height of the panorama, is the initial angle calibration offset between the panoramic device and the spherical camera, which is measured and stored in advance by the device calibration program, when the panoramic device and the spherical camera are at the same angle ; The vertical pitch T combines the vertical field angle ForVer and the pitch angle DangleVer of the ball machine, and a triangular geometry algorithm is used to solve: ； In the formula, the negative sign represents the engineering convention that the downward direction of the ball machine is positive; Zoom factor is: ； In the formula, are the minimum and maximum zoom factors respectively preset for the ball camera.

9. An airport boundary grading early warning linkage system based on fuzzy logic reasoning, characterized in that, The system implements the airport perimeter grading warning linkage method based on fuzzy logic reasoning as claimed in any one of claims 1-8, and the system comprises: A dual-band panoramic monitoring subsystem is used to generate infrared and visible light panoramic images by panoramic scanning of the airport perimeter, and the generated infrared and visible light panoramic image information is sent externally through the MQTT network protocol; An AI intelligent analysis center is used to receive infrared and visible light panoramic image data, and to detect and identify the intrusion target in the perimeter defense area in the infrared and visible light panoramic images through a multi-modal target detection algorithm based on deep learning; the detected and identified intrusion target warning information is pushed, and the intrusion target warning information includes the intrusion target coordinate point and the intrusion target intrusion feature information; An intelligent alarm grading processing center is used to receive and update the sent infrared and visible light panoramic images as the perimeter panoramic base map in real time, to divide the intrusion target warning level through a fuzzy logic algorithm for the sent intrusion target warning information, to judge the priority of the intrusion target warning level, and to push the intrusion target warning level after priority judgment through the HTTP protocol. Linkage control center, for the priority judgment of the intrusion target early warning level of push, using OPENCV algorithm library in the perimeter panoramic base map, the trajectory of the intrusion target is marked, the motion trajectory of the current intrusion target is displayed, and the intention of the intrusion target is judged.

Images