Distributed airport perimeter alarm event identification method and system based on large model

Abstract

The invention belongs to the technical field of civil aviation airport perimeter management, and discloses a distributed airport perimeter alarm event identification method and system based on a large model. According to the method, multiple front-end recognition sensors are comprehensively utilized for comprehensive recognition and judgment, a large model technology is adopted for providing deeper behavior type recognition, airport multi-source data are fused for correction and comparison, and a more intelligent and effective perimeter management technology is provided for airport users. According to the method, multi-party data are fused, multi-source data comparison can be performed, the false alarm rate of the airport compliance business scene is reduced, finally, multiple factors are fused to perform comprehensive alarm event grade evaluation and sorting, and an alarm event processing sequence which should be more concerned by airport users can be provided for the airport users.

Description

Technical Field

[0001] The invention belongs to the technical field of civil aviation airport perimeter management, and in particular relates to a large model-based distributed airport perimeter alarm event recognition method and system. Background Art

[0002] In the traditional airport perimeter management scenario, most airports use perimeter alarm detection technology based on vibration optical cables. In actual applications, the false alarm rate is very high. Thunderstorms, rain, snow, and strong winds may cause a large number of false alarm events, and it is difficult for staff to find effective alarm information in the massive alarm events. In order to solve the high false alarm rate problem caused by a single vibration optical cable technology, some airports have begun to explore adding cameras and millimeter-wave radar recognition technology on the basis of vibration optical cables, intending to use technologies with lower false alarm rates to solve the aforementioned problems. However, in actual applications, due to the lack of effective comprehensive judgment, the false alarm rate has improved compared to before, but it still seriously affects the normal use of staff. At the same time, there is a lack of more fine-grained behavioral discrimination for the identification and judgment of some alarm events. For example, airport security vehicles patrolling around the airport perimeter and gardening construction around the perimeter are often misidentified. In summary, there are many pain points in traditional airport perimeter management, and technical improvements are needed.

[0003] Through the above analysis, the problems and defects of the existing technology are: the traditional airport perimeter recognition technology is single, the false alarm recognition rate is high, and the recognition event types are rare. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the embodiments disclosed in the present invention provide a distributed airport perimeter alarm event recognition method and system based on a large model, specifically a multi-stage multi-factor distributed airport perimeter alarm event recognition and screening method and system based on a large model. The purpose of the present invention is to achieve more accurate perimeter alarm event recognition and sorting by integrating multi-party data and applying multiple algorithm technologies.

[0005] The technical solution is as follows: A distributed airport perimeter alarm event recognition method based on a large model includes: S1: Receive and analyze real-time operation data and historical data, including perimeter alarm identification equipment location data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, meteorological data, and historical alarm record data; S2: Establish an alarm event storage list structure and merge the alarm events; S3: pre-processing the perimeter front-end recognition data and calculating the alarm false alarm rate data; S4: Comprehensively identify multiple devices and environments at the perimeter front end, integrate alarm identification data from millimeter-wave radar, vibration cable, and camera, combine meteorological conditions and historical false alarm rates, and calculate the risk probability of comprehensive target events; S5: The processed alarm events are transmitted to the backend, and before transmission, the transmission data is processed based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factors; S6: Use a multimodal large model trained with targeted fine-tuning for target behavior recognition in alarm events; S7: Calculate the overlap between airport vehicles and alarm events, and calculate the overlap by integrating vehicle dynamic positioning data to identify misjudged alarm events; S8: Calculate the overlap between the farm gardening tasks, perimeter patrol tasks and alarm events; S9: Calculate the current security situation of the airport perimeter and assign different security situation values ​​to alarm events at different alarm identification points; S10: Score the comprehensive priority of the alarm events, and use the value as the basis for the order of different alarm events in the alarm event list displayed to the user.

[0006] In step S1, the parsed perimeter alarm identification equipment point data list is set as , set the parsed machine position data list as , set the parsed public area management gardening task data list as , set the parsed flight operation dynamic data list as , set the parsed vehicle dynamic positioning data list as , set the parsed meteorological data to , set the parsed historical alarm record data list as ; Among them, the number of days of historical data obtained from historical alarm record data Average false alarm rate in recent days , Average number of alarm records per day Average number of flight takeoffs and landings per day Impact, recent time interval to the past sky, The specific value of the day is determined by the user based on the airport's on-site weather conditions and overall operating conditions. , there is false positive data , there is flight takeoff and landing data ; Average number of alarm records per day The calculation method is: Average false alarm rate per day The calculation method is: Average number of flight movements per day The calculation method is: Historical data days The calculation method is: .

[0007] In step S2, an alarm event storage list structure is established to merge the alarm events, including: S201: Establish alarm storage list structure , It is a list structure used to store the alarm events generated by the front end in real time and perform temporary merging calculations; S202: Receive alarm events from front-end devices ,Will Put in ; S203: Create vibration cable alarm event variable , establish millimeter wave radar alarm event variables , create a camera alarm event variable , traverse Data, get the alarm events ; S204: Processing the false alarm data of the vibration optical cable; S205: Create a merged alarm event .

[0008] In step S3, the perimeter front-end recognition data is preprocessed to calculate the alarm false alarm rate data, including: S301: Establishing the false alarm rate of alarm identification points Structure, which is structure, The value stores the alarm identification point number. Store specific false alarm rate data and establish false alarm rate for alarm types structure; S302: Establishing auxiliary calculation of false alarm rate of alarm identification points Structure and structure, The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of identified alarm events of alarm identification points; The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of false alarm events of alarm identification points; establish auxiliary calculation of false alarm rate of alarm type Structure and structure, The structure is structure, The value is used to store the alarm type. Used to store the total number of alarm events identified by the alarm type, The structure is structure, The value is used to store the alarm type. Used to store the total number of false alarm events of the alarm type; S303: Traversal , obtain historical alarm events, obtain the alarm identification point number, alarm type, and whether it is a false alarm data of historical alarm events, and search according to the alarm identification number of historical alarm events ,if If the alarm identification point number does not exist, create a new alarm identification point number. , with 1 as Value data, stored in ; If there is an alarm identification point number, the number The value is increased by 1, The calculation is used to count the number of false positive events of identification numbers; Structure and The structure counts the alarm types; S304: Traversal Structure, get Get the total number of alarm events in the alarm identification number , search by number Get the total number of false positive events in the number , then the false alarm rate of the alarm recognition point The calculation method is: structure is the alarm identification number, To identify the data of point false alarm rate, store ; S305: Traversal Structure, get Get the total number of alarm events in the alarm type , search by type , get the total number of false positive events in the type , then the false alarm rate of the alarm type The calculation method is: structure is the alarm type, For the false alarm rate data, store .

[0009] In step S4, the comprehensive target event risk probability calculation is performed, including: S401: Get , obtain the alarm identification point number and alarm type, and Get the historical false alarm rate of specific alarm identification points , according to the alarm type from Get the historical false alarm rate of specific alarm types ; S402: Obtaining weather data , obtain visibility, rainfall, snow, hail, and wind data, and define visibility data as , visibility unit is meter, and rainfall data is defined as , the snowfall data is defined as , the hail situation data is defined as , the wind condition data is defined as , wind conditions take wind force level data; S403: Calculate the comprehensive risk probability of the alarm event , the expression is: In the formula, is the historical false alarm rate of the alarm identification point of the alarm event, The historical false alarm rate of the alarm event alarm type; S404: Process the alarm event. If the risk probability is ≤50%, delete the alarm event and do not transmit it to the backend. Otherwise, continue with the subsequent steps. S405: Judgment Size, if , then the probability of the alarm event is high, Set the backend processing in "Whether to process in the backend" to "Yes", otherwise to "No".

[0010] In step S5, before transmission, the transmission data is processed based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factor, including: S501: Get alarm events , get the false alarm rate of alarm recognition point , get the false alarm rate of alarm event type , obtain the comprehensive risk probability of the alarm event ; S502: Create a variable to retain the number of images in the camera image , create a variable for the number of millimeter wave radar images to be retained ,but The calculation method is: The calculation method is: S503: Obtaining alarm events Whether the back-end processes the data, if the data is yes, the alarm-related camera data and the alarm-related millimeter-wave radar data are set to empty, and step S504 is skipped, otherwise the subsequent operations are continued; S504: Processing is performed according to the determined number of retained images. The processing method is to traverse the alarm-related camera data and the alarm-related millimeter-wave radar data, delete the even-numbered images therein, and determine whether the number meets the requirement each time an image is deleted. If the requirement is met, the traversal is directly skipped, and if the requirement is not met, the traversal is continued until the number requirement is met; S505: Transmit the processed alarm event data to the backend.

[0011] In step S6, the multimodal large model trained with targeted fine-tuning is used to identify the target behavior of the alarm event, including: S601: Get alarm events , obtain the alarm-related camera data and alarm-related millimeter-wave radar data in the alarm event; S602: Determine the alarm event Whether the backend processes the data, if the data is yes, skip step S603, otherwise continue with the subsequent operations; S603: sequentially merging the images in the alarm-related camera data and the images in the alarm-related millimeter-wave radar data. If the numbers of the two are different, the excess images in the data with a larger number are sequentially merged with the other images from the starting position, and finally a series of camera and millimeter-wave radar integrated images are obtained. S604: The integrated images are sequentially transmitted to the multimodal large model for recognition, and the recognized text is added to the alarm event The alarm text description in; S605: Determine the alarm text description text list added in S604, obtain the alarm type and weather with the highest proportion, and put it into the alarm event Middle and back-end alarm types and meteorological condition attributes.

[0012] In step S7, the overlap between the airport vehicle and the alarm event is calculated, including: S701: Obtain vehicle dynamic positioning data , get alarm events , get alarm events The alarm identification point number, alarm time, and alarm target type; S702: Determine whether the alarm target type is a vehicle. If yes, continue with the subsequent operations. Otherwise, skip S703-S704 operations. S703: Get the list of perimeter alarm identification equipment point data , traverse , obtain the specific perimeter alarm identification device point, compare its number with the alarm identification point number obtained in S701, if the two are consistent, terminate the traversal, otherwise continue the traversal until the perimeter alarm identification device point is found; S704: Obtain the perimeter alarm identification device point data matched in step S702, obtain the latitude and longitude information therein, and traverse , obtain specific vehicle dynamic positioning data; In step S9, the current security situation of the airport perimeter is calculated, and different security situation values ​​are assigned to alarm events at different alarm identification points, including: S901: Get the list of perimeter alarm identification equipment point data , get the list of aircraft position data , get the flight operation dynamic data list as , get alarm events ,create The variable is used to represent the number of flights currently parked at the apron. The initial value is 0. The variable is used to represent the number of flights arriving in the previous hour. The initial value is 0. The variable is used to represent the number of flights departing in the previous hour. The initial value is 0. Create a variable to represent the current time in units of ; S902: Calculate the alarm event and the alarm identification point latitude and longitude data, traverse , obtain the perimeter alarm identification device point data, and compare the number with Compare the alarm identification point numbers in the , if they are consistent, obtain the longitude and latitude data therein, otherwise continue to traverse until the longitude and latitude data that are consistent with the comparison are found; S903: Traversal , obtain specific flight data and determine whether the actual landing time of the flight is If so, then The variable is incremented by 1 to determine whether the actual departure time of the flight is If so, then The variable is incremented by 1. If the flight status is arrival at this airport and not departure at this airport, it is considered that the flight is currently parked on the apron, and the parking space data is obtained. Add 1 to the variable; continue traversing , obtain the aircraft stand data that matches the flight aircraft stand data, and after matching the specific aircraft stand data, obtain the longitude and latitude information therein; S904: Calculate the shortest distance between the longitude and latitude data calculated in S902 and the current parking flight position, and set the shortest distance as a variable , in meters, calculates the distance between the longitude and latitude data calculated in S902 and the longitude and latitude data of the aircraft position currently traversed and obtained in S903 ,if , no operation is performed, otherwise Assigning values , continue to traverse , until the traversal is completed; finally obtain the calculated ; S905: Calculate security situation value , the specific calculation method is as follows: In the formula, Indicates the number of flights currently parked at the apron. Indicates the number of flights arriving in the previous hour. Indicates the number of departing flights in the previous hour. Indicates the shortest distance to the current flight parking position. Adjust parameters for users, who make adjustments as needed.

[0013] S906: The obtained security situation value Endow Risk situation value in ; In step S10, the alarm event comprehensive priority is scored, and the value is used as the order of different alarm events in the alarm event list displayed to the user, and the alarm event priority is set as a variable , specifically including: S1001: Get alarm events , get alarm events Risk probability, meteorological conditions, alarm closed-loop identification, plan matching identification, risk situation value, alarm type, and back-end alarm type data; S1002: Construct backend recognition false positive coefficient variable , indicating the degree of false positives identified by the backend; S1003: Construct backend plan matching identification variable If the plan matching flag is yes, assign 1, otherwise assign 0; construct the back-end alarm closed loop identification flag variable , if the alarm closed loop identification mark is yes, then assign 1, otherwise assign 0; S1004: Construct the front-end identification alarm type priority variable ; S1005: Calculate the alarm event priority score as a variable , the specific calculation method is: in, Indicates the priority value corresponding to the alarm type identified by the front end. Indicates the backend recognition false positive coefficient, Indicates the backend plan matching identifier. Indicates the back-end alarm closed-loop identification mark. represents the risk probability, Indicates the risk situation value.

[0014] S1006: Variables to be calculated Endow The alarm events are added to the alarm event list displayed to the user and reordered based on the priority scores.

[0015] Another object of the present invention is to provide a large-model-based distributed airport perimeter alarm event recognition system, and to implement the large-model-based distributed airport perimeter alarm event recognition method. The large-model-based distributed airport perimeter alarm event recognition system includes: External data collection module: responsible for collecting perimeter alarm identification equipment location data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, and meteorological data; Basic configuration module: responsible for providing basic system operation parameter configuration, basic dictionary data configuration, and user authority configuration functions; Front-end recognition module: responsible for integrating and processing the front-end millimeter-wave radar, camera, and vibration optical cable recognition alarm events, making preliminary judgments and integration on the alarm events, and constructing structured alarm events that meet the requirements of subsequent processing; Front-end comprehensive judgment module: responsible for integrating the alarm recognition data from millimeter-wave radar, vibration cable, and camera, and combining meteorological conditions with historical false alarm rates to conduct a comprehensive target event risk probability analysis, and exclude and filter alarm events with too low risk probabilities; Front-end data transmission processing module: responsible for processing the transmission data based on the false alarm rate of alarm identification points, the false alarm rate of alarm event types, and the comprehensive risk probability factors before transmission, and transmitting the processed data to the back-end for processing; Multimodal recognition module: responsible for identifying the target behavior of alarm events using a multimodal large model that has undergone targeted fine-tuning training; Event closure identification module: responsible for calculating the overlap between the airport vehicle and the alarm event, and calculating the overlap between the airport gardening task, perimeter patrol task plan and the alarm event; Risk situation calculation module: responsible for calculating the current security situation of the airport perimeter and assigning different security situation values ​​to alarm events at different alarm identification points; Comprehensive evaluation and sorting module: responsible for scoring the comprehensive priority of alarm events, sorting the alarm event list displayed to the user, and sending the sorted update data to the interface output module; Interface output module: provides external users with the ability to access the system interface, and provides multiple user access systems via API and Web.

[0016] Combining all the above-mentioned technical solutions, the beneficial effects of the present invention are as follows: the present invention can comprehensively utilize multiple front-end sensor recognition data to perform more accurate front-end alarm event judgment, and at the same time, on the basis of the front-end comprehensive judgment, it provides a back-end generalized behavior understanding and behavior classification based on a large model, which can identify finer-grained and more types of behavioral events. In addition, the present invention integrates multi-party data, can perform multi-source data comparison, and reduce the false alarm rate of airport compliance business scenarios. Finally, the present invention integrates multiple factors to perform comprehensive alarm event level evaluation and ranking, which can provide airport users with the alarm event handling order that they should pay more attention to.

[0017] The present invention solves the problems of high false alarm rate, high false alarm quantity and coarse recognition granularity in traditional airport perimeter solutions.

[0018] The present invention solves the technical problems of high false alarm rate, high number of false alarms and rough recognition granularity existing in the traditional airport perimeter technical solutions. In the traditional technical solutions, vibration optical cable or millimeter wave radar and other technologies are used for single-stage recognition. Due to the problems of the recognition hardware itself and the failure to fully consider the on-site operating environment and situation, a high false alarm rate and number of false alarms occur. The present invention first integrates the recognition data of various recognition hardware in the front-end recognition, and considers the operating environment and the historical false alarm rate at the same time, so as to give more accurate recognition data at the front-end recognition. In the subsequent multi-modal large model recognition, it makes up for the shortcoming of low recognition accuracy of the traditional front-end recognition technology for occasional small sample events. At the same time, it integrates multi-source system data, comprehensively calculates the airport security situation, and then gives the importance ranking of the recognized security events, which reduces the work cost of repeated clicks to confirm in the traditional technical solutions, and improves the overall airport security management and control level. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure; Figure 1 It is a flow chart of a distributed airport perimeter alarm event identification method based on a large model provided by an embodiment of the present invention; Figure 2 It is a flow chart of merging and processing alarm events provided by an embodiment of the present invention; Figure 3 is a flow chart of calculating the false alarm rate of an alarm provided by an embodiment of the present invention; Figure 4 is a flow chart of comprehensive target event risk probability calculation provided by an embodiment of the present invention; Figure 5 is a flowchart of alarm data transmission processing provided by an embodiment of the present invention; Figure 6 It is a flow chart of target behavior recognition of multi-modal large model alarm events provided by an embodiment of the present invention; Figure 7 is a flow chart of calculating the coincidence degree between airport vehicles and alarm events provided by an embodiment of the present invention; Figure 8 It is a flow chart of calculating the coincidence of farm gardening tasks, perimeter patrol task plans and alarm events provided by an embodiment of the present invention; Fig. 9 is a flow chart of the current security situation of the perimeter of a computer farm provided by an embodiment of the present invention; Fig.10 It is a flow chart of calculating the comprehensive priority score of alarm events provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0021] The overall innovation of the present invention is that in response to the problems of high false alarm rate, high number of false alarms and rough recognition granularity existing in traditional perimeter management technology, the present invention comprehensively weighs multiple factors such as multiple front-end recognition equipment recognition, weather, historical false alarm rate, etc., and proposes a more accurate front-end alarm event discrimination, and adopts a large model technology that has been fine-tuned for targeted scenarios to provide deeper and more fine-grained behavior type recognition, and integrates multi-source data of the airport to calculate data overlap and the real-time safe operation status of the airport, providing users with a comprehensive consideration of the airport's historical operating rules, real-time operating status, environmental changes, and multiple sensors. The alarm event sorting scheme provides a more intelligent perimeter management technology.

[0022] Embodiment 1, as Figure 1 As shown, the embodiment of the present invention provides a distributed airport perimeter alarm event identification method based on a large model, including: S1: Receive and analyze real-time operation data and historical data, including perimeter alarm identification equipment location data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, meteorological data, and historical alarm record data; The data includes perimeter alarm identification equipment point data, aircraft position data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, meteorological data, and historical alarm record data. The perimeter alarm identification equipment point data includes the alarm identification point number, longitude and latitude, identification equipment, update and maintenance time, and creation time. The aircraft position data includes the aircraft position number, aircraft position category, longitude and latitude, regional attributes, corridor bridge position identification, and self-sliding identification. The public area management gardening task data includes the task number, task name, task status, task plan start time, task actual start time, task plan end time, and task actual end time. The security management system perimeter patrol task data includes the task number, task name, task status, task plan start time, task actual start time, task plan end time, and task actual end time. Logical task data includes task number, task name, task status, task-related vehicles, task planned start time, task actual start time, task planned end time, and task actual end time. Flight movement dynamic data includes flight number, inbound flight number, outbound flight number, task attributes, flight status, flight abnormal status, route, VIP identification, key task type, and aircraft position. Vehicle dynamic positioning data includes vehicle number, license plate number, longitude and latitude, speed, and update time. Meteorological data includes number, visibility, rainfall, snowfall, hail, and wind conditions. Historical alarm record data includes number, alarm identification point number, alarm time, alarm type, alarm description, whether it is a false alarm, and handling suggestions.

[0023] Set the parsed perimeter alarm identification equipment point data list as , set the parsed machine position data list as , set the parsed public area management gardening task data list as , set the parsed flight operation dynamic data list as , set the parsed vehicle dynamic positioning data list as , set the parsed meteorological data to , set the parsed historical alarm record data list as ; Among them, the number of days of historical data obtained from historical alarm record data Average false alarm rate in recent days , Average number of alarm records per day Average number of flight takeoffs and landings per day Impact, recent time interval to the past sky, The specific value of the day is determined by the user based on the airport's on-site weather conditions and overall operating conditions. , there is false positive data , there is flight takeoff and landing data ; Average number of alarm records per day The calculation method is: Average false alarm rate per day The calculation method is: Average number of flight movements per day The calculation method is: Historical data days The calculation method is: S2: Establish an alarm event storage list structure and merge the alarm events; S201: Establish alarm storage list structure , It is a list structure used to store the alarm events generated by the front end in real time and perform temporary merging calculations; S202: Receive alarm events from front-end devices ,Will Put in ; The alarm event includes the alarm identification point number, alarm device number, device type, alarm type, backend alarm type, alarm target type, alarm-related camera data, alarm-related millimeter-wave radar data, alarm time, risk probability, risk importance score, whether backend processing, alarm text description, weather conditions, alarm closed-loop identification mark, plan matching mark, risk situation value, and priority score. This step receives the alarm event. The most original event, at the same time, the same point containing millimeter-wave radar, vibration cable, and camera equipment may generate 0-3 different identification device type events. The alarm-related camera data will store 10 related images, and the associated millimeter-wave radar data will store 8 related images. Put in ; S203: Create vibration cable alarm event variable , establish millimeter wave radar alarm event variables , create a camera alarm event variable , traverse Data, get the alarm events ; Will The alarm event is represented as , defining the current time as , the allowed difference between the alarm event alarm time and the current time is defined as , It is a user-adjustable parameter. Users can adjust it according to the operation needs. First, determine , if not more than Then continue to traverse If more than Then terminate this traversal and obtain If the device type is vibration cable, Data endowment If the device type is millimeter wave radar, Data endowment , if the device type is a camera, then Data endowment , get The alarm identification point number is based on the number in Search in to obtain the alarm event with the same alarm identification point number, and judge the newly obtained alarm event If it exceeds , then assign the corresponding aforementioned variables according to the newly acquired alarm event device type; S204: Processing the false alarm data of the vibration optical cable; if The variable is not null, and Variables and If the variable is null, the vibration cable data is considered a false alarm, and no merged alarm event is created. Variable from Delete from; S205: Create a merged alarm event ; Structure and To be consistent, will not be empty Data consolidation is given, where The device type is a list structure. Alarm-associated camera data and alarm-associated millimeter-wave radar data respectively store camera-associated data and millimeter-wave radar-associated data in the alarm event.

[0024] For example, Figure 2 The embodiment of the present invention provides a flow chart for merging and processing alarm events, including: 1) establishing an alarm storage structure (corresponding to S201); 2) receiving front-end alarm events (corresponding to S202); 3) alarm event preprocessing (corresponding to S203); 4) vibration optical cable false alarm processing (corresponding to S204); 5) determining whether it is a false alarm, if so, merging and processing the alarm events ends; if not, creating a merged alarm event, until merging and processing the alarm events ends (corresponding to S204 and S205).

[0025] S3: pre-processing the perimeter front-end recognition data and calculating the alarm false alarm rate data; like Figure 3 As shown, the specific steps are as follows: S301: Establishing the false alarm rate of alarm identification points Structure, which is structure, The value stores the alarm identification point number. Store specific false alarm rate data and establish false alarm rate for alarm types structure; the structure is structure, The value is used to store the alarm type. The specific false alarm rate data is stored. The present invention can identify the alarm types as shown in Table 1: Table 1 Identifiable alarm types S302: Establishing auxiliary calculation of false alarm rate of alarm identification points Structure and structure, First, an auxiliary calculation structure is established to calculate the number of warnings and false alarms at each identification point in the historical data, as well as the number of warnings and false alarms for each alarm type, and then the final calculation of the false alarm rate is performed based on these data; The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of identified alarm events of alarm identification points; The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of false alarm events of alarm identification points; establish auxiliary calculation of false alarm rate of alarm type Structure and structure, The structure is structure, The value is used to store the alarm type. Used to store the total number of alarm events identified by the alarm type, The structure is structure, The value is used to store the alarm type. Used to store the total number of false alarm events of the alarm type; S303: Traversal , obtain historical alarm events, obtain the alarm identification point number, alarm type, and whether it is a false alarm data of historical alarm events, and search according to the alarm identification number of historical alarm events ,if If the alarm identification point number does not exist, create a new alarm identification point number. , with 1 as Value data, stored in ; If there is an alarm identification point number, the number The value is increased by 1, The calculation is used to count the number of false positive events of identification numbers; Structure and The structure counts the alarm types; S304: Traversal Structure, get Get the total number of alarm events in the alarm identification number , search by number Get the total number of false positive events in the number , then the false alarm rate of the alarm recognition point The calculation method is: structure is the alarm identification number, To identify the data of point false alarm rate, store ; S305: Traversal Structure, get Get the total number of alarm events in the alarm type , search by type , get the total number of false positive events in the type , then the false alarm rate of the alarm type The calculation method is: structure is the alarm type, For the false alarm rate data, store .

[0026] S4: Perform comprehensive identification of multiple devices and environments at the perimeter front end, integrate alarm identification data from millimeter wave radar, vibration cable, and camera, combine meteorological conditions and historical false alarm rates, and calculate the risk probability of comprehensive target events; Figure 4 As shown, the specific calculation steps are as follows: S401: Get , obtain the alarm identification point number and alarm type, and Get the historical false alarm rate of specific alarm identification points , according to the alarm type from Get the historical false alarm rate of a specific alarm type ; The rainfall conditions are divided into 0-10 levels, 0 means no rainfall, and the larger the value, the heavier the rainfall. The snowfall data is defined as The snowfall condition is divided into 0-10 levels, 0 means no snowfall, and the larger the value, the heavier the snowfall. The hail condition data is defined as The hail situation is divided into 0-10 levels, 0 means no hail, and the larger the value, the heavier the hail. The wind situation data is defined as , wind conditions take wind force level data; S402: Obtaining weather data , obtain visibility, rainfall, snow, hail, and wind data, and define visibility data as , visibility unit is meter, and rainfall data is defined as , the snowfall data is defined as , the hail situation data is defined as , the wind condition data is defined as , wind conditions take wind force level data; S403: Calculate the comprehensive risk probability of the alarm event , the expression is: In the formula, is the historical false alarm rate of the alarm identification point of the alarm event, The historical false alarm rate of the alarm event alarm type; S404: Process the alarm event. If the risk probability is ≤50%, delete the alarm event and do not transmit it to the backend. Otherwise, continue with the subsequent steps. S405: Judgment Size, if , then the probability of the alarm event is high, Set the backend processing in "Whether to process in the backend" to "Yes", otherwise to "No".

[0027] S5: The processed alarm events are transmitted to the backend, and before transmission, the transmission data is processed based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factors; In order to reduce network bandwidth consumption while taking into account recognition accuracy, such as Figure 5 As shown, the specific steps are as follows: S501: Get alarm events , get the false alarm rate of alarm recognition point , get the false alarm rate of alarm event type , obtain the comprehensive risk probability of the alarm event ; S502: Create a variable to retain the number of images in the camera image , create a variable for the number of millimeter wave radar images to be retained ,but The calculation method is: The calculation method is: S503: Obtaining alarm events Whether the back-end processes the data, if the data is yes, the alarm-related camera data and the alarm-related millimeter-wave radar data are set to empty, and step S504 is skipped, otherwise the subsequent operations are continued; S504: Processing is performed according to the determined number of retained images. The processing method is to traverse the alarm-related camera data and the alarm-related millimeter-wave radar data, delete the even-numbered images therein, and determine whether the number meets the requirement each time an image is deleted. If the requirement is met, the traversal is directly skipped, and if the requirement is not met, the traversal is continued until the number requirement is met; S505: Transmit the processed alarm event data to the backend.

[0028] S6: Use a multimodal large model trained with targeted fine-tuning for target behavior recognition in alarm events; First, a large model that has been fine-tuned and trained using a multimodal base model is constructed. Specifically, fine-tuning data is first constructed. In order to ensure that the subsequent fine-tuned multimodal large model can meet the output requirements, the present invention constructs the following types of data as shown in Table 2; Table 2 Type data Among them, gardening work, internal patrols, internal operation and maintenance, agricultural labor, and airport employees are common types of false alarm events, which are not easy to identify at the front end. They can be supplemented by using a multimodal large model for identification; behavioral anomalies supplement the front-end alarm events with additional abnormal behavior information; abnormal weather identifies the weather and can be compared and corrected with the received weather data.

[0029] The processing of multimodal image data in the present invention is different from conventional data processing methods. The camera images and millimeter-wave radar images of historically collected alarm events are spliced ​​and integrated, and fine-tuned as a complete image for training, so that the multimodal large model can have a more comprehensive understanding of the environmental conditions.

[0030] Exemplarily, the text structure of the present invention for an image adopts the structural form of "{{event detailed description}}{{event type}}{{weather conditions}}", and the specific example is as follows: {{Airport property company gardening staff are pruning vegetation near the perimeter}}{{gardening work}}{{rainfall}}.

[0031] After fine-tuning the multimodal large model, the present invention will use the multimodal large model to identify and process the data in the alarm event. The specific steps are as follows: S601: Get alarm events , obtain the alarm-related camera data and alarm-related millimeter-wave radar data in the alarm event; S602: Determine the alarm event Whether the backend processes the data, if the data is yes, skip step S603, otherwise continue with the subsequent operations; S603: sequentially merging the images in the alarm-related camera data and the images in the alarm-related millimeter-wave radar data. If the numbers of the two are different, the excess images in the data with a larger number are sequentially merged with the other images from the starting position, and finally a series of camera and millimeter-wave radar integrated images are obtained. S604: The integrated images are sequentially transmitted to the multimodal large model for recognition, and the recognized text is added to the alarm event The alarm text description in; S605: Determine the alarm text description text list added in S604, obtain the alarm type and weather with the highest proportion, and put it into the alarm event Middle and back-end alarm types and meteorological condition attributes.

[0032] S7: Calculate the overlap between airport vehicles and alarm events, and calculate the overlap by integrating vehicle dynamic positioning data to identify misjudged alarm events; like Figure 7 As shown, the specific steps are as follows: S701: Obtain vehicle dynamic positioning data , get alarm events , get alarm events The alarm identification point number, alarm time, and alarm target type; S702: Determine whether the alarm target type is a vehicle. If yes, continue with the subsequent operations. Otherwise, skip S703-S704 operations. S703: Get the list of perimeter alarm identification equipment point data , traverse , obtain the specific perimeter alarm identification device point, compare its number with the alarm identification point number obtained in S701, if the two are consistent, terminate the traversal, otherwise continue the traversal until the perimeter alarm identification device point is found; S704: Obtain the perimeter alarm identification device point data matched in step S702, obtain the latitude and longitude information therein, and traverse , obtain specific vehicle dynamic positioning data; S8: Calculate the overlap between the farm gardening tasks, perimeter patrol tasks and alarm events; like Figure 8 As shown, the specific steps are as follows: S801: Obtaining public area management gardening task data , security management system perimeter patrol mission data , get alarm events , get the alarm time in the alarm event ; S802: Traversal and , get the task data, if the task actual start time is not empty, then the task start time is the actual start time of the task, otherwise the task start time Schedule the start time and end time of the task The acquisition method is similar; S803: Determine whether and If so, then Set the plan matching flag in to yes, otherwise continue traversing and , until the traversal ends.

[0033] S9: Calculate the current security situation of the airport perimeter and assign different security situation values ​​to alarm events at different alarm identification points; like Fig. 9 As shown, the specific steps are as follows: S901: Get the list of perimeter alarm identification equipment point data , get the list of aircraft position data , get the flight operation dynamic data list as , get alarm events ,create The variable is used to represent the number of flights currently parked at the apron. The initial value is 0. The variable is used to represent the number of flights arriving in the previous hour. The initial value is 0. The variable is used to represent the number of flights departing in the previous hour. The initial value is 0. Create a variable to represent the current time in units of ; S902: Calculate the alarm event and the alarm identification point latitude and longitude data, traverse , obtain the perimeter alarm identification device point data, and compare the number with Compare the alarm identification point numbers in the , if they are consistent, obtain the longitude and latitude data therein, otherwise continue to traverse until the longitude and latitude data that are consistent with the comparison are found; S903: Traversal , obtain specific flight data and determine whether the actual landing time of the flight is If so, then The variable is incremented by 1 to determine whether the actual departure time of the flight is If so, then The variable is incremented by 1. If the flight status is arrival at this airport and not departure at this airport, it is considered that the flight is currently parked on the apron, and the parking space data is obtained. Add 1 to the variable; continue traversing , obtain the aircraft stand data that matches the flight aircraft stand data, and after matching the specific aircraft stand data, obtain the longitude and latitude information therein; S904: Calculate the shortest distance between the longitude and latitude data calculated in S902 and the current parking flight position, and set the shortest distance as a variable , in meters, calculates the distance between the longitude and latitude data calculated in S902 and the longitude and latitude data of the aircraft position currently traversed and obtained in S903 ,if , no operation is performed, otherwise Assigning values , continue to traverse , until the traversal is completed; finally obtain the calculated ; S905: Calculate security situation value , the specific calculation method is as follows: In the formula, Indicates the number of flights currently parked at the apron. Indicates the number of flights arriving in the previous hour. Indicates the number of departing flights in the previous hour. Indicates the shortest distance to the current flight parking position. Adjust parameters for users, who make adjustments as needed.

[0034] S906: The obtained security situation value Endow Risk situation value in ; S10: Score the comprehensive priority of the alarm events, and use the value as the basis for the order of different alarm events in the alarm event list displayed to the user.

[0035] S1001: Get alarm events , get alarm events Risk probability, meteorological conditions, alarm closed-loop identification, plan matching identification, risk situation value, alarm type, and back-end alarm type data; S1002: Construct backend recognition false positive coefficient variable , indicating the degree of false positives identified by the backend; Determined by the backend alarm type data. See Table 3 for specific value selection rules.

[0036] Table 3 Value rules S1003: Construct backend plan matching identification variable If the plan matching flag is yes, assign 1, otherwise assign 0; construct the back-end alarm closed loop identification flag variable , if the alarm closed loop identification mark is yes, then assign 1, otherwise assign 0; refer to the above operation, the risk probability is a variable , the risk situation value is a variable .

[0037] S1004: Construct the front-end identification alarm type priority variable , as shown in Table 4; Table 4 Alarm types and priorities S1005: Calculate the alarm event priority score as a variable , the specific calculation method is: in, Indicates the priority value corresponding to the alarm type identified by the front end. Indicates the backend recognition false positive coefficient, Indicates the backend plan matching identifier. Indicates the back-end alarm closed-loop identification mark. represents the risk probability, Indicates the risk situation value.

[0038] S1006: Variables to be calculated Endow The alarm events are added to the alarm event list displayed to the user and reordered based on the priority scores.

[0039] For example, Fig.10The present invention provides a comprehensive priority scoring process for calculating alarm events, including: 1) obtaining alarm event attribute data (corresponding to S1001); 2) constructing a calculation backend identification false alarm coefficient (corresponding to S1002); 3) constructing a calculation backend plan matching identification variable (corresponding to S1003); 4) constructing a calculation front-end identification alarm type priority variable (corresponding to S1004); 5) calculating the alarm event priority score (corresponding to S1005); 6) assigning and sorting the alarm event priority score (corresponding to S1006).

[0040] Embodiment 2, the distributed airport perimeter alarm event recognition system based on the large model provided by the embodiment of the present invention comprises an external data acquisition module, a basic configuration module, a front-end recognition module, a front-end comprehensive judgment module, a front-end data transmission processing module, a multimodal recognition module, an event closure recognition module, a risk situation calculation module, a comprehensive evaluation ranking module, and an interface output module; the specific functions and interrelationships of each module are as follows: External data collection module: responsible for collecting perimeter alarm identification equipment point data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, meteorological data, etc.

[0041] Basic configuration module: responsible for providing functions such as basic system operation parameter configuration, basic dictionary data configuration, and user authority configuration.

[0042] Front-end recognition module: responsible for integrating and processing the front-end millimeter-wave radar, camera, and vibration optical cable recognition alarm events, making preliminary judgments and integration on the alarm events, and constructing structured alarm events that meet the subsequent processing requirements.

[0043] Front-end comprehensive judgment module: responsible for integrating the alarm identification data from millimeter-wave radar, vibration cable, and camera, combining meteorological conditions and historical false alarm rates, to conduct a comprehensive target event risk probability, and exclude and filter alarm events with too low risk probabilities.

[0044] Front-end data transmission processing module: responsible for processing the transmission data based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factors before transmission, and transmitting the processed data to the back-end for processing.

[0045] Multimodal recognition module: responsible for identifying the target behavior of alarm events on the multimodal large model that has undergone targeted fine-tuning training, and is also responsible for providing support for the stable operation of the multimodal large model.

[0046] Event closure identification module: responsible for calculating the overlap between airport vehicles and alarm events, and calculating the overlap between airport gardening tasks, perimeter patrol task plans and alarm events, providing a data basis for comprehensive evaluation and sorting.

[0047] Risk situation calculation module: responsible for calculating the current security situation of the airport perimeter, assigning different security situation values ​​to alarm events at different alarm identification points, and providing a data basis for comprehensive evaluation and ranking.

[0048] Comprehensive evaluation and sorting module: responsible for scoring the comprehensive priority of alarm events, sorting the alarm event list displayed to the user, and sending the sorted update data to the interface output module.

[0049] Interface output module: The module provides external users with the ability to access the system interface, provides multiple user access system forms such as API and Web, and can support the input of alarm event data into other related systems.

[0050] It can be seen from the above embodiments that the present invention proposes a multi-stage, multi-factor distributed airport perimeter alarm event identification and screening method and system based on a large model. In response to the problems of high false alarm rate, high number of false alarms, and rough identification granularity in traditional perimeter management technology, the present invention comprehensively weighs multiple factors such as multiple front-end identification equipment identification, weather, and historical false alarm rates, and proposes a more accurate front-end alarm event discrimination. At the same time, compared with the single-stage discrimination of traditional perimeter technology, the present invention adopts a multi-stage comprehensive discrimination technology, uses a multi-modal large model to improve the fine-grained behavior recognition capability, comprehensively considers the calculation of multi-data overlap, and calculates the security situation in combination with the real-time operation of the airport, and provides a comprehensive alarm event sorting scheme to reduce the false alarm rate of airport compliance business scenarios, improve the efficiency of users in handling alarm events, and thus improve the overall perimeter safety management level of the airport.

[0051] In order to effectively compare the differences between the method and system of the present invention and the traditional technical solutions, the overall identification process of perimeter alarm events is now compared in stages. According to the technical execution flow analysis, perimeter alarm event identification can be divided into front-end device identification triggering, front-end data processing and transmission, back-end comprehensive judgment, and data processing push stages. In the front-end device identification triggering stage, the traditional technical solutions mainly use vibration optical cables in identification hardware, or some airports use millimeter-wave radar and camera recognition technology. After identifying the perimeter alarm, it will directly enter the next stage. The method of the present invention integrates vibration optical cables, millimeter-wave radars, and camera recognition data, and considers environmental conditions and the historical false alarm rate of the identification node. It evaluates the importance of alarm events and removes obvious false alarm events. Comprehensively comparing this stage, the method of the present invention will be able to obtain more accurate alarm events and reduce false alarms to a certain extent. In the front-end data processing and transmission stage, the traditional technical solution mainly structures the alarm events identified by the front-end and then transmits them to the back-end. The present invention will perform transmission data processing based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factor, so as to reduce the network bandwidth consumption on the basis of taking into account the recognition accuracy. In this stage, the present invention will transmit additional images and radar data on the basis of transmitting structured alarm events. At the same time, in order to reduce the bandwidth, a comprehensive processing method is adopted, and the additional data will be more conducive to the judgment of the back-end. In the back-end comprehensive judgment stage, the traditional technical solution basically has no additional operation, and transmits the alarm event data identified by the front-end to the next stage, while the method of the present invention will use a multi-modal large model to improve the fine-grained behavior recognition capability, comprehensively consider the calculation of multi-data overlap, and combine the real-time operation of the airport to calculate the security situation. In this stage, the present invention will be able to add a deeper level of behavior recognition capability to the alarm event, and at the same time, it can integrate the actual operation of the airport and the security situation, and can tell the user which alarm events are more critical in a more comprehensive way. In the data processing and pushing stage, the traditional technical solution will structure the front-end identified alarm events and pass the processed alarm event data to the front-end page, while the present invention will evaluate and calculate the previous alarm events and give the alarm event priority ranking, which can allow users to focus more on important and urgent security event processing. Overall, compared with the traditional technical solution, the present invention has obvious technical implementation effect advantages and can significantly improve the user's perimeter security management level.

Claims

1. A distributed airport perimeter alarm event recognition method based on a large model, characterized in that: The method includes: S1: Receive and analyze real-time operation data and historical data, including perimeter alarm identification equipment location data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, meteorological data, and historical alarm record data; S2: Establish an alarm event storage list structure and merge the alarm events; S3: pre-processing the perimeter front-end recognition data and calculating the alarm false alarm rate data; S4: Perform comprehensive identification of multiple devices and environments at the perimeter front end, integrate alarm identification data from millimeter wave radar, vibration cable, and camera, combine meteorological conditions and historical false alarm rates, and calculate the risk probability of comprehensive target events; S5: The processed alarm events are transmitted to the backend, and before transmission, the transmission data is processed based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factors; S6: Use a multimodal large model trained with targeted fine-tuning for target behavior recognition in alarm events; S7: Calculate the overlap between airport vehicles and alarm events, and calculate the overlap by integrating vehicle dynamic positioning data to identify misjudged alarm events; S8: Calculate the overlap between the farm gardening tasks, perimeter patrol tasks and alarm events; S9: Calculate the current security situation of the airport perimeter and assign different security situation values ​​to alarm events at different alarm identification points; S10: Score the comprehensive priority of the alarm events, and use the value as the basis for the order of different alarm events in the alarm event list displayed to the user.

2. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S1, the parsed perimeter alarm identification equipment point data list is set as , set the parsed machine position data list as , set the parsed public area management gardening task data list as , set the parsed flight operation dynamic data list as , set the parsed vehicle dynamic positioning data list as , set the parsed meteorological data to , set the parsed historical alarm record data list as ; Among them, the number of days of historical data obtained from historical alarm record data Average false alarm rate in recent days , Average number of alarm records per day Average number of flight takeoffs and landings per day Impact, recent time interval to past sky, The specific value of the day is determined by the user based on the airport's on-site weather conditions and overall operating conditions. , there is false positive data , there is flight takeoff and landing data ; Average number of alarm records per day The calculation method is: Average false alarm rate per day The calculation method is: Average number of flight movements per day The calculation method is: Historical data days The calculation method is: 。 3. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S2, an alarm event storage list structure is established to merge the alarm events, including: S201: Establish alarm storage list structure , It is a list structure used to store the alarm events generated by the front end in real time and perform temporary merging calculations; S202: Receive alarm events from front-end devices ,Will Put in ; S203: Create vibration cable alarm event variable , establish millimeter wave radar alarm event variables , create a camera alarm event variable , traverse Data, get the alarm events ; S204: Processing the false alarm data of the vibration optical cable; S205: Create a merged alarm event .

4. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S3, the perimeter front-end recognition data is preprocessed to calculate the alarm false alarm rate data, including: S301: Establishing the false alarm rate of alarm identification points Structure, which is structure, The value stores the alarm identification point number. Store specific false alarm rate data and establish false alarm rate for alarm types structure; S302: Establishing auxiliary calculation of false alarm rate of alarm identification points Structure and structure, The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of identified alarm events of alarm identification points; The structure is structure, The value is used to store the alarm identification point number. Used to store the total number of false alarm events of alarm identification points; establish auxiliary calculation of false alarm rate of alarm type Structure and structure, The structure is structure, The value is used to store the alarm type. Used to store the total number of alarm events identified by the alarm type, The structure is structure, The value is used to store the alarm type. Used to store the total number of false alarm events of the alarm type; S303: Traversal , obtain historical alarm events, obtain the alarm identification point number, alarm type, and whether it is a false alarm data of historical alarm events, and search according to the alarm identification number of historical alarm events ,if If the alarm identification point number does not exist, create a new alarm identification point number. , with 1 as Value data, stored in ; If there is an alarm identification point number, the number The value is increased by 1, The calculation is used to count the number of false positive events of identification numbers; Structure and The structure counts the alarm types; S304: Traversal Structure, get Get the total number of alarm events in the alarm identification number , search by number Get the total number of false positive events in the number , then the false alarm rate of the alarm recognition point The calculation method is: structure is the alarm identification number, To identify the data of point false alarm rate, store ; S305: Traversal Structure, get Get the total number of alarm events in the alarm type , search by type , get the total number of false positive events in the type , then the false alarm rate of the alarm type The calculation method is: structure is the alarm type, For the false alarm rate data, store .

5. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S4, the comprehensive target event risk probability calculation is performed, including: S401: Get , obtain the alarm identification point number and alarm type, and Get the historical false alarm rate of specific alarm identification points , according to the alarm type from Get the historical false alarm rate of specific alarm types ; S402: Obtaining weather data , obtain visibility, rainfall, snow, hail, and wind data, and define visibility data as , visibility unit is meter, and rainfall data is defined as , the snowfall data is defined as , the hail situation data is defined as , the wind condition data is defined as , wind conditions take wind force level data; S403: Calculate the comprehensive risk probability of the alarm event , the expression is: In the formula, is the historical false alarm rate of the alarm identification point of the alarm event, The historical false alarm rate of the alarm event alarm type; S404: Process the alarm event. If the risk probability is ≤50%, delete the alarm event and do not transmit it to the backend. Otherwise, continue with the subsequent steps. S405: Judgment Size, if , then the probability of the alarm event is high, Set the backend processing in "Whether to process in the backend" to "Yes", otherwise to "No".

6. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S5, before transmission, the transmission data is processed based on the false alarm rate of the alarm identification point, the false alarm rate of the alarm event type, and the comprehensive risk probability factor, including: S501: Get alarm events , get the false alarm rate of alarm recognition point , get the false alarm rate of alarm event type , obtain the comprehensive risk probability of the alarm event ; S502: Create a variable to retain the number of images in the camera image , create a variable for the number of millimeter wave radar images to be retained ,but The calculation method is: The calculation method is: S503: Obtaining alarm events Whether the back-end processes the data, if the data is yes, the alarm-related camera data and the alarm-related millimeter-wave radar data are set to empty, and step S504 is skipped, otherwise the subsequent operations are continued; S504: Processing is performed according to the determined number of retained images. The processing method is to traverse the alarm-related camera data and the alarm-related millimeter-wave radar data, delete the even-numbered images therein, and determine whether the number meets the requirement each time an image is deleted. If the requirement is met, the traversal is directly skipped, and if the requirement is not met, the traversal is continued until the number requirement is met; S505: Transmit the processed alarm event data to the backend.

7. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S6, the multimodal large model trained with targeted fine-tuning is used to identify the target behavior of the alarm event, including: S601: Get alarm events , obtain the alarm-related camera data and alarm-related millimeter-wave radar data in the alarm event; S602: Determine the alarm event Whether the backend processes the data, if the data is yes, skip step S603, otherwise continue with the subsequent operations; S603: sequentially merging the images in the alarm-related camera data and the images in the alarm-related millimeter-wave radar data. If the numbers of the two are different, the excess images in the data with a larger number are sequentially merged with the other images from the starting position, and finally a series of camera and millimeter-wave radar integrated images are obtained. S604: The integrated images are sequentially transmitted to the multimodal large model for recognition, and the recognized text is added to the alarm event The alarm text description in; S605: Determine the alarm text description text list added in S604, obtain the alarm type and weather with the highest proportion, and put it into the alarm event Middle and back-end alarm types and meteorological condition attributes.

8. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S7, the overlap between the airport vehicle and the alarm event is calculated, including: S701: Obtain vehicle dynamic positioning data , get alarm events , get alarm events The alarm identification point number, alarm time, and alarm target type; S702: Determine whether the alarm target type is a vehicle. If yes, continue with the subsequent operations. Otherwise, skip S703-S704 operations. S703: Get the list of perimeter alarm identification equipment point data , traverse , obtain the specific perimeter alarm identification device point, compare its number with the alarm identification point number obtained in S701, if the two are consistent, terminate the traversal, otherwise continue the traversal until the perimeter alarm identification device point is found; S704: Obtain the perimeter alarm identification device point data matched in step S702, obtain the latitude and longitude information therein, and traverse , obtain specific vehicle dynamic positioning data; In step S8: calculating the overlap between the gardening task, perimeter patrol task plan and the alarm event includes: S801: Obtaining public area management gardening task data , security management system perimeter patrol mission data , get alarm events , get the alarm time in the alarm event ; S802: Traversal and , get the task data, if the task actual start time is not empty, then the task start time is the actual start time of the task, otherwise the task start time Schedule the start time and end time of the task The acquisition method is similar; S803: Determine whether and If so, then Set the plan matching flag in to yes, otherwise continue traversing and , until the traversal ends.

9. The large model-based distributed airport perimeter alarm event recognition method according to claim 1 is characterized in that: In step S9, the current security situation of the airport perimeter is calculated, and different security situation values ​​are assigned to alarm events at different alarm identification points, including: S901: Get the list of perimeter alarm identification equipment point data , get the list of aircraft position data , get the flight operation dynamic data list as , get alarm events ,create The variable is used to represent the number of flights currently parked at the apron. The initial value is 0. The variable is used to represent the number of flights arriving in the previous hour. The initial value is 0. The variable is used to represent the number of flights departing in the previous hour. The initial value is 0. Create a variable to represent the current time in units of ; S902: Calculate the alarm event and the alarm identification point latitude and longitude data, traverse , obtain the perimeter alarm identification device point data, and compare the number with Compare the alarm identification point numbers in the , if they are consistent, obtain the longitude and latitude data therein, otherwise continue to traverse until the longitude and latitude data that are consistent with the comparison are found; S903: Traversal , obtain specific flight data and determine whether the actual landing time of the flight is If so, then The variable is incremented by 1 to determine whether the actual departure time of the flight is If so, then The variable is incremented by 1. If the flight status is arrival at this airport and not departure at this airport, it is considered that the flight is currently parked on the apron, and the parking space data is obtained. Add 1 to the variable; continue traversing , obtain the aircraft stand data that matches the flight aircraft stand data, and after matching the specific aircraft stand data, obtain the longitude and latitude information therein; S904: Calculate the shortest distance between the longitude and latitude data calculated in S902 and the current parking flight position, and set the shortest distance as a variable , in meters, calculates the distance between the longitude and latitude data calculated in S902 and the longitude and latitude data of the aircraft position currently traversed and obtained in S903 ,if , no operation is performed, otherwise Assigning values , continue to traverse , until the traversal is completed; finally obtain the calculated ; S905: Calculate security situation value , the specific calculation method is as follows: In the formula, Indicates the number of flights currently parked at the apron. Indicates the number of flights arriving in the previous hour. Indicates the number of departing flights in the previous hour. Indicates the shortest distance to the current flight parking position. Adjust parameters for users, who make adjustments as needed; S906: The obtained security situation value Endow Risk situation value in ; In step S10, the alarm event comprehensive priority is scored, and the value is used as the order of different alarm events in the alarm event list displayed to the user, and the alarm event priority is set as a variable , specifically including: S1001: Get alarm events , get alarm events Risk probability, meteorological conditions, alarm closed-loop identification, plan matching identification, risk situation value, alarm type, and back-end alarm type data; S1002: Construct backend recognition false alarm coefficient variable , indicating the degree of false positives identified by the backend; S1003: Construct backend plan matching identification variable If the plan matching flag is yes, assign 1, otherwise assign 0; construct the back-end alarm closed loop identification flag variable , if the alarm closed loop identification mark is yes, then assign 1, otherwise assign 0; S1004: Construct the front-end identification alarm type priority variable S1005: Calculate the alarm event priority score as a variable , the specific calculation method is: in, Indicates the priority value corresponding to the alarm type identified by the front end. Indicates the backend recognition false positive coefficient, Indicates the backend plan matching identifier. Indicates the back-end alarm closed-loop identification mark. represents the risk probability, Indicates the risk situation value; S1006: Variables to be calculated Endow The alarm events are added to the alarm event list displayed to the user and re-ordered based on the priority scores.

10. A distributed airport perimeter alarm event recognition system based on a large model, characterized in that: Implementing the large model-based distributed airport perimeter alarm event recognition method as described in any one of claims 1 to 9, the large model-based distributed airport perimeter alarm event recognition system comprises: External data collection module: responsible for collecting perimeter alarm identification equipment location data, aircraft station location data, public area management gardening task data, security management system perimeter patrol task data, flight operation dynamic data, vehicle dynamic positioning data, and meteorological data; Basic configuration module: responsible for providing basic system operation parameter configuration, basic dictionary data configuration, and user authority configuration functions; Front-end recognition module: responsible for integrating and processing the front-end millimeter-wave radar, camera, and vibration optical cable recognition alarm events, making preliminary judgments and integration on the alarm events, and constructing structured alarm events that meet the requirements of subsequent processing; Front-end comprehensive judgment module: responsible for integrating the alarm recognition data from millimeter-wave radar, vibration cable, and camera, and combining meteorological conditions with historical false alarm rates to conduct a comprehensive target event risk probability analysis, and exclude and filter alarm events with too low risk probabilities; Front-end data transmission processing module: responsible for processing the transmission data based on the false alarm rate of alarm identification points, the false alarm rate of alarm event types, and the comprehensive risk probability factors before transmission, and transmitting the processed data to the back-end for processing; Multimodal recognition module: responsible for identifying the target behavior of alarm events using a multimodal large model that has undergone targeted fine-tuning training; Event closure identification module: responsible for calculating the overlap between the airport vehicle and the alarm event, and calculating the overlap between the airport gardening task, perimeter patrol task plan and the alarm event; Risk situation calculation module: responsible for calculating the current security situation of the airport perimeter and assigning different security situation values ​​to alarm events at different alarm identification points; Comprehensive evaluation and sorting module: responsible for scoring the comprehensive priority of alarm events, sorting the alarm event list displayed to the user, and sending the sorted update data to the interface output module; Interface output module: provides external users with the ability to access the system interface, and provides multiple user access systems via API and Web.

Images