Pipeline detection method for civil aviation guarantee

Abstract

The application discloses a pipeline detection method for civil aviation guarantee, comprising the following steps: adopting a target detection algorithm to analyze a standardized video frame sequence, identifying personnel and equipment positions involved in a refueling process, and determining position coordinate data; if the position coordinate data indicates that the personnel or equipment deviates from a preset safety area, analyzing the standardized video frame sequence through a behavior classification model to determine whether an abnormal behavior exists, and obtaining a behavior classification result; sending a risk event label and a corresponding standardized video frame sequence segment to a monitoring center through a data stream transmission technology to generate an alarm signal; recording the alarm signal and the standardized video frame sequence segment through an all-weather data storage technology to generate a data record; analyzing key events in the refueling process according to the data record and a real-time operation state extraction technology, adjusting an operation process through a dynamic scheduling algorithm, optimizing refueling task scheduling, and obtaining a scheduling instruction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and specifically relates to a pipeline detection method for civil aviation guarantee. BACKGROUND

[0002] In the process of aircraft refueling, real-time acquisition, analysis and application of dynamic visual information by means of high-resolution cameras and image processing technology is an important link to ensure the safe and efficient operation of civil aviation. However, this process faces many core technical problems.

[0003] First, in the high-dynamic and high-complexity refueling scene, how to ensure the integrity, real-time and accuracy of visual data is the primary problem. The refueling operation environment is complex and changeable, personnel and equipment move at high speed, light conditions are unstable, visibility differs greatly under different weather conditions, and sudden abnormal behaviors may occur, all of which bring great challenges to data acquisition.

[0004] Secondly, in terms of data processing, high-frame-rate cameras will capture a large amount of video data, and how to quickly process and store these data through image compression and edge computing technology to ensure that the data stream is not interrupted or lost due to high load is a problem to be solved. If data processing is not timely or storage is not proper, key information may be lost, affecting subsequent analysis and decision-making.

[0005] Thirdly, the accuracy of the target detection algorithm is crucial. The algorithm needs to accurately identify the position coordinates of personnel and equipment, and when it detects deviation from the safety area, it needs to quickly determine the nature of the abnormal behavior through a behavior classification model. However, existing algorithms may misjudge or miss due to environmental noise or dynamic occlusion, thereby causing safety hazards.

[0006] In addition, compliance verification based on image segmentation technology and standard operation templates needs to deal with diversified operation actions and equipment form differences to ensure the reliability of the determination results. Different operation habits of different operators and differences in equipment models may affect the accuracy of compliance verification.

[0007] At the same time, real-time data stream transmission to the monitoring center needs to overcome network delay and bandwidth limitation to ensure the synchronicity and integrity of risk event labels and video segments. If there is a problem in the transmission process, the monitoring center may not be able to obtain key information in time, delaying the handling of risk events.

[0008] All-weather data storage system needs to cope with the long-term storage and fast retrieval of large-scale video data. As time goes on, the amount of data increases, and how to efficiently store and quickly retrieve data is crucial for event tracing and analysis.

[0009] Finally, the dynamic scheduling algorithm needs to handle optimization problems under multiple variable constraints when analyzing data records and real-time states to generate accurate refueling task scheduling instructions. Refueling tasks involve multiple factors such as equipment status, personnel arrangement, time constraints, etc. The algorithm needs to consider these factors comprehensively to achieve optimal scheduling.

[0010] The above problems are all around the core problem of efficient collection and intelligent analysis of visual data in dynamic and complex scenarios, involving comprehensive challenges in data processing, algorithm accuracy, and system collaboration. SUMMARY

[0011] The purpose of the present application is to provide a pipeline detection method for civil aviation support, to solve the problem of how to realize efficient collection and intelligent analysis of visual data in the dynamic and complex scenario of aircraft refueling, as proposed in the background art, to address the comprehensive technical challenges of data collection integrity and real-time deficiency, data processing and storage pressure, low target detection and compliance verification accuracy, real-time transmission and storage limitations, and dynamic scheduling optimization difficulties.

[0012] To solve the above technical problems, the technical solution adopted by the present application is:

[0013] A pipeline detection method for civil aviation support, comprising the following steps:

[0014] Step S101, real-time photographing of the aircraft refueling process is performed by a high-resolution camera and image processing technology, and the image data is converted into a digital format to form standardized video frames that can be analyzed;

[0015] Step S102, a target detection algorithm is used to analyze the standardized video frame sequence, identify the positions of personnel and equipment involved in the refueling process, and determine the position coordinate data;

[0016] Step S103, if the position coordinate data indicates that personnel or equipment deviates from the pre-set safety area, a behavior classification model is used to analyze the standardized video frame sequence to determine whether abnormal behavior exists, and obtain the behavior classification result;

[0017] Step S104, automatically identify whether the operation conforms to the standard process using a visual analysis algorithm, and mark a risk event tag for operations that do not conform to the standard process;

[0018] Step S105, send the risk event tag and the corresponding standardized video frame sequence segment to the monitoring center through data streaming technology to generate an alarm signal;

[0019] Step S106, record the alarm signal and standardized video frame sequence segment using all-weather data storage technology to generate a data record;

[0020] Step S107, according to the data record and real-time operation state extraction technology, analyze the key events in the refueling process, adopt dynamic scheduling algorithm to adjust the operation process, optimize the refueling task scheduling, and get the scheduling instruction.

[0021] According to the above technical scheme, in step S101, the high-resolution camera and image processing technology are used to take pictures of the aircraft refueling process in real time and convert the image data into digital format to form an analyzable visual data stream, including:

[0022] The high-frame-rate camera captures the dynamic scene of aircraft refueling to obtain a continuous image sequence;

[0023] The continuous image sequence is compressed and denoised, and is converted into digital visual data when the definition condition is met;

[0024] The digital visual data is stored, indexed and key frame feature extracted to generate an analyzable visual data stream.

[0025] According to the above technical scheme, in step S102, the target detection algorithm is used to analyze the standardized video frame sequence, identify the positions of personnel and equipment involved in the refueling process, and determine the position coordinate data, including:

[0026] The target detection algorithm is used to identify the positions of personnel and equipment, and generate position coordinate data;

[0027] Based on the position coordinate data, a trajectory data set is constructed, and the motion pattern features are extracted after smoothing processing;

[0028] Combined with the preset rule library, the abnormal operation is judged, the video frame sequence is labeled and a structured event data stream is generated.

[0029] According to the above technical scheme, in step S103, if the position coordinate data indicates that the personnel or equipment deviates from the preset safety area, the behavior classification model is used to analyze the standardized video frame sequence to determine whether the abnormal behavior exists, and the behavior classification result is obtained, including:

[0030] When the position coordinate data exceeds the boundary of the preset safety area, it is determined that the personnel or equipment position deviates;

[0031] Extract the video frame subset of the deviation period, and use the convolutional neural network to analyze and extract the behavior pattern features;

[0032] Combined with the preset behavior rule library, the abnormal behavior is judged, the corresponding video frame is labeled, and a structured abnormal event data stream is generated, and the boundary of the preset safety area is updated.

[0033] According to the technical scheme, in step S104, the operation whether it conforms to the standard process is automatically identified by combining the visual analysis algorithm, and the operation not conforming to the standard process is marked with a risk event label, including:

[0034] The video frame sequence is processed by an image segmentation technology to separate the refueling equipment and the operation personnel action;

[0035] The operation personnel action feature is extracted, an action mode vector is generated, and the action mode vector is compared with a standard operation template database;

[0036] The non-compliant action is identified and labeled, a structured non-compliant event data stream is generated, and an operation quality evaluation database is updated.

[0037] According to the technical scheme, in step S105, the risk event label and the corresponding standardized video frame sequence segment are sent to the monitoring center by a data stream transmission technology to generate an alarm signal, including:

[0038] The video frame segment labeled with the risk event label is acquired, a standardized video frame data stream is generated, and the standardized video frame data stream is sent to the monitoring center;

[0039] When the number of risk event labels exceeds a preset threshold, an alarm signal data is generated, and a high-priority video segment is extracted;

[0040] The high-priority video segment is processed to generate structured event data, a risk event database of the monitoring center is updated, and a dynamic alarm signal is determined.

[0041] According to the technical scheme, in step S106, all-weather data storage technology is used to record the alarm signal and the standardized video frame sequence segment to generate a data record, including:

[0042] The alarm signal and the video frame sequence are collected to form an original data set, and the video frame sequence is standardized;

[0043] If the frame rate is not up to standard, missing frames are supplemented, and the complete frame sequence is time-stamped aligned with the alarm signal;

[0044] The synchronous data set is analyzed to determine an event trigger point, relevant data is extracted to generate a data record, and the data record is compressed and stored.

[0045] According to the technical scheme, in step S107, according to the data record and real-time operation state extraction technology, a key event in the refueling process is analyzed, a dynamic scheduling algorithm is used to adjust the operation process, a refueling task scheduling is optimized, a scheduling instruction is obtained, including:

[0046] The key event is extracted from the data record and real-time state data and classified to determine a priority;

[0047] If there is a high-priority event, the operation flow is adjusted by a dynamic programming algorithm to generate an optimized task scheduling scheme;

[0048] Verify the validity of the scheduling parameters, generate scheduling instructions and adjust the operation sequence in real time, and feed back the updated dynamic scheduling algorithm.

[0049] According to the above technical scheme, it also includes:

[0050] Extract key events from data records and system state, determine their timestamps and priorities;

[0051] When the priority of the key event is higher than the preset threshold, use a heuristic scheduling algorithm to sort the tasks and adjust the allocation scheme;

[0052] Verify and update the task instructions in combination with the real-time system state, generate the final scheduling instructions and verify their reliability.

[0053] According to the above technical scheme, it also includes:

[0054] Parse the data records to obtain historical operation data, extract the key event set and determine the initial sequence of task allocation in combination with the real-time state;

[0055] If the initial sequence meets the scheduling constraint condition, optimize the task sequence, otherwise update the key event set;

[0056] Generate scheduling instructions and verify their executability, generate feedback data according to the execution deviation, update the scheduling algorithm parameters and key event extraction rules.

[0057] Compared with the prior art, the present application has the following beneficial effects:

[0058] In the present application, the dynamic operation scene is captured by a high-resolution high-frame-rate camera, combined with an image compression algorithm and an edge computing device, realizing real-time processing and storage of data, constructing a visual data acquisition system supporting an event database, ensuring efficient acquisition, digitization of dynamic visual information, and data integrity and real-time performance; And using target detection algorithm to accurately identify personnel and equipment position coordinates, combined with behavior classification model can quickly judge abnormal behavior, improve the identification ability of safety risk, reduce the possibility of misjudgment and omission.

[0059] Then use image segmentation technology and compare with the standard operation template database to realize automatic verification of operation compliance, provide reliable basis for operation quality evaluation, and help to standardize the refueling operation process; risk events are transmitted to the monitoring center through data flow and generate alarm signals, use all-weather data storage technology to record data records, facilitate timely processing of risk events and event tracing.

[0060] In terms of task scheduling, based on data records and real-time states, a dynamic scheduling algorithm is used to optimize the refueling task scheduling, accurate scheduling instructions are generated, the scheduling efficiency of the refueling process is significantly improved, and the operation safety and quality are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A flowchart of the pipeline detection method for civil aviation guarantee according to the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0063] Embodiment one

[0064] As shown in the figure, the present embodiment provides a pipeline detection method for civil aviation guarantee, mainly including: Figure 1

[0065] S101: visual data acquisition and digital processing.

[0066] Through high-resolution cameras and image processing technology, the aircraft refueling process is photographed in real time and the image data is converted into digital format to form an analyzable visual data stream to support event recording and subsequent intelligent analysis. The business problem focuses on how to efficiently collect and digitize dynamic visual information in the refueling process, ensure data integrity and real-time performance, and the technical details include using high-frame-rate cameras to capture dynamic operation scenes, combining image compression algorithms and edge computing devices to realize fast data processing and storage, and building a basic visual data acquisition system supporting event databases.

[0067] The specific steps are: capturing the aircraft refueling dynamic scene by a high-frame-rate camera to obtain a continuous image sequence; using a JPEG compression algorithm to compress the continuous image sequence to obtain compressed image data; performing a denoising operation on the compressed image data on an edge computing device to generate denoised image data; if the clarity of the denoised image data is higher than a preset threshold, converting it into digital visual data; storing the digital visual data to the local cache through the edge computing device to generate a temporary data stream; constructing a visual data index of the event database according to the timestamp sequence of the temporary data stream to obtain structured data; extracting key frame features from the structured data and storing them to the event database to generate an analyzable visual data stream.

[0068] ​Specifically, by deploying high-resolution cameras, such as 4K industrial cameras equipped with 50 million pixel CMOS sensors and supporting 120 frames per second, real-time capture of dynamic scenes during aircraft refueling is ensured, capturing key operation details such as refueling pipe connection, generating high-definition image streams. The camera is installed 5 meters above the refueling area, covering a 10m x 10m field of view, combined with automatic focusing and wide-angle lenses to ensure complete recording of dynamic scenes. Image data is processed by H.265 compression algorithm, with a compression ratio of 50:1, compressing about 20MB of raw data per frame to about 400KB, reducing bandwidth requirements while retaining over 90% visual details. The compressed data stream is transmitted to the edge computing device, which selects the NVIDIA Jetson AGX Orin module with a computing power of 200 trillion operations per second, running real-time image processing algorithms such as YOLOv8 target detection model, identifying key components such as refueling gun position and fuel tank interface, with a detection accuracy of 95%, and processing time per frame controlled within 10 milliseconds, ensuring real-time performance. The edge device uploads the processed data to the cloud event database at a rate of 100Mbps through a 5G network, the database uses PostgreSQL combined with the time series plugin TimescaleDB to store structured visual data, inserting 1000 records per second, including timestamps, image feature vectors and event labels. Data integrity is verified by SHA-256 hash check, generating a unique hash value for each frame of data to ensure no loss or tampering during transmission. The system supports subsequent intelligent analysis, such as extracting refueling operation anomaly features through convolutional neural networks (CNN), analyzing oil leaks or operation errors, with an anomaly detection accuracy of 98%. To support business association, the system integrates multiple sensor data synchronously with visual data, combined with timestamps for cross-modal analysis, generating refueling efficiency reports to optimize scheduling decisions.

[0069] Further, the generated visual data stream for analysis will serve as the basis for subsequent S102 target detection, S103 behavior analysis and S104 compliance identification. After the edge computing device completes processing, only key feature data (such as structured coordinates, anomaly labels) is uploaded to the cloud, and the original video frame sequence is retained in local edge storage, ensuring data transmission efficiency and real-time performance.

[0070] S102: Target detection and position coordinate determination.

[0071] The standardized video frame sequence is analyzed by using a target detection algorithm to identify the positions of personnel and equipment involved in the refueling process and determine position coordinate data. The position coordinate data is used for space-time mapping to construct trajectories of the personnel and equipment in the refueling scene and generate a trajectory dataset. If the continuity of the trajectory points in the trajectory dataset is higher than a preset threshold, the trajectory is smoothed to obtain smoothed trajectory data. Motion patterns of the personnel and equipment in the refueling process are extracted from the smoothed trajectory data to generate motion pattern features. The motion pattern features are matched with a preset rule library to determine whether there is an abnormal operation and generate an abnormality determination result. The video frame sequence is labeled by using the abnormality determination result to generate video frame data with abnormality labels. An event association index is constructed by using the video frame data with abnormality labels to generate a structured event data stream.

[0072] The specific steps are as follows: the standardized video frame sequence is analyzed by using a target detection algorithm to identify the positions of personnel and equipment involved in the refueling process and generate position coordinate data; the position coordinate data is used for space-time mapping to construct trajectories of the personnel and equipment in the refueling scene and generate a trajectory dataset; if the continuity of the trajectory points in the trajectory dataset is higher than a preset threshold, the trajectory is smoothed to obtain smoothed trajectory data; motion patterns of the personnel and equipment in the refueling process are extracted from the smoothed trajectory data to generate motion pattern features; the motion pattern features are matched with a preset rule library to determine whether there is an abnormal operation and generate an abnormality determination result; the video frame sequence is labeled by using the abnormality determination result to generate video frame data with abnormality labels; an event association index is constructed by using the video frame data with abnormality labels to generate a structured event data stream.

[0073] Specifically, by deploying a multispectral camera, the dynamic positions of personnel and equipment during aircraft refueling are captured, generating a video frame sequence containing infrared and visible light information, with a resolution of 2560x1440 pixels and 60 frames per second, ensuring clear images of personnel and equipment under different lighting conditions. The video frame sequence is input into a target detection algorithm based on CenterNet, using ResNet-50 as the backbone network to extract feature maps of personnel and equipment (such as refueling trucks and hoses), generating bounding boxes and center point coordinates for each object, with a coordinate accuracy of ±0.1 meters and a processing time of 15 milliseconds per frame. The detection algorithm runs on an edge computing device, using an Intel Movidius Myriad X chip with a processing capacity of 1 trillion operations per second, supporting real-time analysis. After detection, the coordinate data is structured in JSON format, including object type (personnel or equipment), timestamp, and three-dimensional coordinates (x, y, z), generating 500 records per second, which are transmitted to a local Redis cache database with a storage delay of less than 5 milliseconds. To ensure data consistency, a CRC32 checksum algorithm is used to generate a 32-bit checksum for each set of coordinate data, verifying that the data is lossless during transmission and storage. The coordinate data is further smoothed by a Kalman filter algorithm to predict the motion trajectory of personnel and equipment, reducing coordinate jumps caused by occlusion or noise, with a prediction error of less than 0.05 meters. The processed coordinate data is combined with environmental sensor data (such as wind speed and temperature) from the refueling station, and through time series analysis, a position distribution heat map of personnel and equipment is generated, stored in a MongoDB database, supporting subsequent scheduling optimization and safety monitoring, with a heat map resolution of 100x100 grid, representing a 0.1 square meter area for each grid.

[0074] Further, the target detection algorithm of this step runs on an edge computing device, based on the structured data generated in S101 for real-time analysis, and the position coordinate data is simultaneously pushed to the spatial analysis module of S103 and the real-time state database of S107, providing spatial dimension basis for safety area judgment and scheduling optimization.

[0075] S103: Abnormal behavior judgment.

[0076] If the position coordinate data indicates that personnel or equipment deviates from the pre-set safety area, the standardized video frame sequence is analyzed by a behavior classification model to determine whether abnormal behavior exists, obtaining a behavior classification result.

[0077] The specific steps are: if the position coordinate data exceeds the preset safety region boundary, comparing the position coordinate data with the safety region boundary through the spatial analysis module to determine the personnel position deviation or the equipment position deviation; according to the personnel position deviation or the equipment position deviation, extracting a video frame subset corresponding to a time period from the standardized video frame sequence to obtain a deviation period frame sequence; using a convolutional neural network to analyze the deviation period frame sequence to extract behavior pattern features to obtain a behavior pattern dataset; if the behavior pattern dataset does not match the preset behavior rule library, judging the abnormal behavior by comparing the behavior pattern features with the abnormal detection threshold to obtain an abnormal behavior identifier; according to the abnormal behavior identifier, labeling the corresponding frames from the video frame sequence to generate video frame data with an abnormal label; constructing a time sequence index through the video frame data with an abnormal label to generate a structured abnormal event data stream; using the structured abnormal event data stream to update the preset safety region boundary to obtain optimized safety region data.

[0078] Specifically, based on the position coordinate data, when it is detected that the personnel or the equipment deviates from the preset safety region, for example, a safety boundary of 5 meters from a refueling point), the system automatically triggers the behavior classification model to analyze the standardized video frame sequence. The video frame sequence is captured by a high-resolution camera with a resolution of 1920x1080 pixels, 30 frames per second, covering the full panorama of the refueling area. The frame sequence is first input into the behavior classification model based on YOLOv5, which uses a CSPDarknet53 backbone network to extract behavior features of personnel and equipment, such as the walking posture of personnel and the moving speed of equipment. The model is trained using a labeled dataset containing 10,000 samples, covering normal behaviors (such as regular refueling operations) and abnormal behaviors (such as personnel running quickly or equipment deviating abnormally), with a classification accuracy of 95%. After feature extraction, the model generates a behavior classification probability for each frame, classified into normal and abnormal categories, with a processing time controlled within 20 milliseconds. The classification results are combined with the coordinate data to further analyze whether the behavior constitutes a safety risk through a logistic regression algorithm, for example, if the personnel deviates from the safety region by more than 1 meter within 2 seconds or the equipment moves at a speed exceeding 0.5 meters per second. The analysis results generate JSON data containing behavior type, probability value (0 to 1), and timestamp, with 300 records generated per second, stored in a local Elasticsearch database with an index delay of less than 10 milliseconds. To ensure classification stability, the system uses a sliding window mechanism with a window size of 5 frames to calculate the mean value of the behavior probability, reducing false positives caused by single-frame noise. Finally, the abnormal behavior data is analyzed through time series analysis to generate a behavior anomaly distribution map with a resolution of 50x50 grid, with each grid representing a 0.2 square meter area, stored in a Cassandra database to support real-time safety alerts and subsequent behavior audits.

[0079] Further, this step is a parallel analysis process with S104, and the structured abnormal event data stream will be one of the core sources of S105 risk event labels (with higher priority than compliance issues), while triggering the high-priority event response mechanism of S107.

[0080] S104: Operation compliance identification.

[0081] Automatically identify whether the operation conforms to the standard process combined with visual analysis algorithms, and focus on how to automatically verify the compliance of refueling operations through visual data. The technical details include using image segmentation technology to identify refueling equipment and operator actions, comparing with the standard operation template database, generating compliance judgment results, and supporting operation quality evaluation.

[0082] The specific steps are: processing the video frame sequence of the gas station by image segmentation technology, separating the refueling equipment and the operator's actions, and obtaining the segmented image data; using a preset standard process template to extract the operator's action features from the segmented image data to obtain an action feature data set; analyzing the action feature data set through a convolutional neural network to generate an action pattern vector; if the action pattern vector deviates from the preset vector in the standard operation template database by more than a preset threshold, it is determined to be a non-compliant action, and a non-compliant action identifier is obtained; according to the non-compliant action identifier, the corresponding frames are labeled from the video frame sequence to generate video frame data with non-compliant labels; constructing a time series index through the video frame data with non-compliant labels to generate a structured non-compliant event data stream; using the structured non-compliant event data stream to update the operation quality evaluation database to obtain optimized evaluation data.

[0083] Specifically, the system captures the video stream of the refueling area through a high-resolution camera, with a resolution of 1280x720 pixels and 25 frames per second, covering the core operating area of the gas station. The video frames are first input into a Mask R-CNN-based image segmentation algorithm. The model uses a ResNet-50 backbone network, pre-trained on the COCO dataset, and fine-tuned for the refueling scenario, with a segmentation accuracy of 92%. The algorithm identifies the refueling gun, tank mouth, and operator's arm in real time, generating a pixel-level mask for each object. For example, the mask area of the refueling gun should be maintained between 5000 and 8000 pixels, and the center point coordinate error of the tank mouth should be less than 10 pixels. The segmentation results are compared with a standard operation template database, which contains 3000 standard refueling action sequences, covering refueling gun insertion, removal, and holding angle, etc. The duration of each action in the template ranges from 2 to 5 seconds, and the angle deviation is less than 15 degrees. The comparison process uses a dynamic time warping algorithm to calculate the similarity between the spatiotemporal sequence of the segmentation mask and the template, with a score range of 0 to 1, and a threshold of 0.85. If the score is below this threshold, it is determined to be non-compliant. The system further extracts action features, such as the average curvature of the arm movement trajectory (less than 0.3) and the refueling gun insertion speed (controlled within 0.2 to 0.4 meters per second), and judges whether the action meets the standard through a support vector machine classifier. The analysis results are generated in JSON format, including action type, similarity score, and timestamp, with 200 records generated per second, stored in a MongoDB database, and an index delay of less than 8 milliseconds. To improve robustness, the system uses a 3-frame sliding window to calculate the average similarity, reducing false positives caused by changes in lighting or occlusion. Finally, the compliance determination results are transmitted to the monitoring system through Kafka streaming, supporting real-time operation quality evaluation and subsequent auditing. The audit data is stored in a 100x100 grid format, with each grid representing a 0.1 square meter area, recording the frequency and location of non-compliant actions.

[0084] Further, this step is independent of S103 and monitors the compliance of the entire refueling process regardless of whether the personnel / equipment deviates from the safety area. The structured non-compliant event data stream will be used as a supplementary source (with lower priority than abnormal behavior) for S105 risk event tagging. When the cumulative number of non-compliant actions exceeds the pre-set threshold, the scheduling adjustment mechanism of S107 is triggered simultaneously.

[0085] S105: Alarm signal generation.

[0086] The risk event tags and corresponding standardized video frame sequences are sent to the monitoring center through data streaming technology, generating an alarm signal.

[0087] Specific steps are: through a data transmission protocol, obtaining a video frame segment labeled with a risk event label from a video frame sequence, generating a standardized video frame data stream; using a streaming data transmission technology, sending the standardized video frame data stream to a monitoring center to obtain real-time transmission data; if the number of risk event labels in the real-time transmission data exceeds a preset threshold, generating an alarm signal data through a signal generation rule; according to the alarm signal data, extracting the corresponding video frame segment from the standardized video frame data stream to obtain a high-priority video segment; through a data stream processing technology, performing time sequence indexing on the high-priority video segment to generate structured event data; using the structured event data, updating a risk event database of the monitoring center to obtain an optimized database record; through the optimized database record, generating a real-time data stream to determine a dynamic alarm signal of the monitoring center.

[0088] Specifically, the system captures video data through high-definition cameras deployed at gas stations, with a resolution of 1920x1080 pixels and 30 frames per second, covering an area of 5x5 meters around the fueling area to generate continuous video streams. The video streams are first subjected to target detection using the YOLOv5 algorithm, with a CSPDarkNet-53 backbone network pre-trained on the VOC dataset and transferred to the fueling scenario for learning. The detection accuracy reaches 95%, enabling real-time identification of fueling guns, tank openings, and abnormal objects (such as oil spills or foreign objects). The detection results generate bounding box data with an area between 3000 and 10000 pixels and a center point coordinate error within 8 pixels. The system then performs spatiotemporal sequence analysis on the detected targets, using an LSTM network to extract motion features such as the displacement speed of the fueling gun (between 0.3 and 0.5 meters per second) and the relative position change of the tank opening (horizontal offset less than 5 centimeters). The analysis results generate JSON data containing target type, motion speed, and timestamp, with 150 records generated per second, subjected to streaming processing using Apache Flink with a data transmission delay of less than 10 milliseconds. The processed data are compared with a risk event template library containing 2000 abnormal event sequences, such as incorrect insertion of fueling guns or oil spills, with a cosine similarity algorithm used in the matching process and a similarity threshold set at 0.9. A value below this threshold triggers a risk label. The label data and corresponding video frame segments (5 seconds each, containing 150 frames) are transmitted at a rate of 500 MB per second to the monitoring center through Kafka to generate an alarm signal containing the risk type, location coordinates, and timestamp, stored in a Redis database with a query delay of less than 5 milliseconds. To ensure robustness, the system uses a 5-frame sliding window to calculate the mean of the motion features, filtering out noise interference. The alarm signal is pushed to the monitoring center interface in JSON format, supporting real-time risk visualization and subsequent event tracing, with data stored in a 50x50 grid representing a 0.2 square meter area and recording the frequency of risk events.

[0089] Further, the priority rules of the risk event label are: the abnormal behavior label of S103 (priority 1) > the non-compliance action cumulative label of S104 (priority 2). During transmission, the 5G slicing technology is used to guarantee the transmission bandwidth of the high-priority video segment, and to ensure that the monitoring center receives the alarm signal and the associated video segment within 3 seconds.

[0090] S106: Data record generation.

[0091] The alarm signal and the standardized video frame sequence segment are recorded by using all-weather data storage technology to generate a data record.

[0092] The specific steps are: collecting the alarm signal and the video frame sequence through the all-weather storage system to obtain an original data set; using standardized processing to convert the video frame sequence in the original data set to generate a standardized frame sequence; if the frame rate in the standardized frame sequence is lower than a preset threshold, then missing frames are supplemented by using an interpolation algorithm to obtain a complete frame sequence; time stamp alignment is performed on the complete frame sequence and the alarm signal to determine a synchronous data set; the alarm signal and the frame sequence in the synchronous data set are analyzed by using an event detection algorithm to determine an event trigger point; the associated frame sequence segment and the alarm signal are extracted according to the event trigger point to generate a data record; the data record is stored by using a data compression technology to obtain compressed data.

[0093] Further, the all-weather storage system of this step and the event database of S101 are the same architecture, and use an "edge local storage + cloud backup" mode: the edge device stores the original video frames and data records for nearly 72 hours, and the data exceeding 72 hours is automatically compressed and uploaded to the cloud cold storage to ensure the data traceability and the balance between storage efficiency.

[0094] S107: Dispatching instruction generation.

[0095] According to the data record and real-time operation state extraction technology, the key events in the refueling process are analyzed, a dynamic scheduling algorithm is used to adjust the operation process, the refueling task scheduling is optimized, and a dispatching instruction is obtained.

[0096] The specific steps are: obtaining original data of the refueling process from data records and real-time operation state data, extracting key events through time series analysis to obtain a key event set; classifying events using a clustering algorithm according to the key event set to determine an event priority sequence; if there is a high-priority event in the event priority sequence, adjusting the operation process through a dynamic programming algorithm to generate an optimized task scheduling scheme; extracting scheduling parameters from the optimized task scheduling scheme, verifying parameter effectiveness through real-time state monitoring to obtain a verification result; if the verification result meets a preset threshold, generating scheduling instructions according to the scheduling parameters to output an instruction set; adjusting the refueling process in real time according to the instruction set using time window analysis technology to obtain an adjusted operation sequence; extracting execution logs from the adjusted operation sequence, updating the dynamic scheduling algorithm through feedback analysis to obtain an optimized algorithm model. Specifically, the original data of the refueling process is obtained from data records and real-time operation state data, key events are extracted through time series analysis to obtain a key event set; events are classified using a clustering algorithm according to the key event set to determine an event priority sequence; if there is a high-priority event in the event priority sequence, the operation process is adjusted through a dynamic programming algorithm to generate an optimized task scheduling scheme; scheduling parameters are extracted from the optimized task scheduling scheme, parameter effectiveness is verified through real-time state monitoring to obtain a verification result; if the verification result meets a preset threshold, scheduling instructions are generated according to the scheduling parameters to output an instruction set; the refueling process is adjusted in real time according to the instruction set using time window analysis technology to obtain an adjusted operation sequence; execution logs are extracted from the adjusted operation sequence, the dynamic scheduling algorithm is updated through feedback analysis to obtain an optimized algorithm model.

[0097] Specifically, based on data records and real-time operation state extraction technology, key events in the refueling process can be analyzed through log analysis and sensor data collection.

[0098] By analyzing data records and real-time state to extract key events, a dynamic scheduling algorithm is used to adjust the operation process and task allocation to obtain optimized refueling task scheduling instructions.

[0099] The raw data stream is obtained from the data records and system state, the key events are extracted by parsing technology, the timestamp and priority of the events are determined. If the priority of the key event is higher than the preset threshold, a heuristic scheduling algorithm is used to prioritize the tasks to obtain an optimized task sequence. According to the optimized task sequence, the task allocation scheme is adjusted in combination with the system resource utilization to generate a preliminary refueling task instruction. Through real-time data processing, the dynamic changes of the current system state are obtained, and it is judged whether the task instruction needs to be adjusted. If the task instruction does not match the current system state, the heuristic scheduling algorithm is re-run, the task allocation scheme is updated, and a new refueling task instruction is obtained. According to the updated task allocation scheme, in combination with the event priority sorting, the final refueling task scheduling instruction is generated. Through instruction generation accuracy verification, the rule engine is used to compare the expected results to determine the reliability of the final instruction.

[0100] Specifically, by analyzing data records and real-time state to extract key events, log parsing tools can be used in combination with time series analysis to process historical and real-time data. By analyzing data records and real-time state, key events are extracted, and a dynamic scheduling algorithm is used to adjust the operation process and task allocation to determine the optimized refueling task scheduling instruction.

[0101] By parsing data records, historical operation data and timestamps are obtained from the storage system, events containing refueling task trigger conditions are extracted, and a set of key events is obtained. According to the set of key events, in combination with real-time state monitoring data, the device running state and task priority are analyzed to determine the initial sequence of task allocation. If the initial sequence meets the preset scheduling constraint condition, a heuristic algorithm is used to adjust the task allocation to generate an optimized task sequence; if not, the real-time state is re-parsed and the set of key events is updated. Through the optimized task sequence, in combination with device availability and time window, the scheduling instruction of the refueling task is generated. According to the scheduling instruction, the executability of the instruction is verified, if the instruction does not match the real-time state, the heuristic algorithm is re-executed, the task sequence is adjusted, and the updated scheduling instruction is obtained. The updated scheduling instruction is obtained, in combination with historical execution data, the deviation of instruction execution is analyzed, and feedback data is generated for subsequent optimization and adjustment. Through the feedback data, the parameters of the scheduling algorithm are updated, new key event extraction rules are generated, and the generation of the next task scheduling instruction is optimized.

[0102] For example, based on data records and real-time operation state, key events are extracted, such as a refueling task being paused due to pressure abnormalities. Time series analysis is used to calculate the duration of the abnormality and the loss, a genetic algorithm is used to optimize the operation process, the task priority, device state and time constraints are input, and the optimal scheduling scheme is generated after 100 iterations. The scheduling instruction is sent to the control system through the API interface, and is associated with inventory management to ensure the feasibility of the task, improve refueling efficiency, and shorten waiting time.

[0103] Further, by analyzing data records and real-time state, key events are extracted, dynamic scheduling algorithm is used to adjust operation process and task allocation, and optimized refueling task scheduling instructions are obtained. The original data stream is obtained from the data records and system state, the timestamp and priority of the key event are extracted, if the priority of the key event is higher than the preset threshold, the heuristic scheduling algorithm is used to sort the task, the allocation scheme is adjusted to generate the preliminary instruction, the real-time data processing is used to judge whether the instruction needs to be adjusted, if it does not match, the algorithm is re-run to update the instruction, the final instruction is generated by combining the event priority sorting, and the reliability is verified by the rule engine.

[0104] Further, the historical operation data and timestamp are obtained by analyzing the data records, and the key event set is extracted; the initial sequence of task allocation is determined by combining the real-time state monitoring data; if the initial sequence meets the scheduling constraint condition, the heuristic algorithm is used to optimize the task sequence, otherwise the key event set is updated; the scheduling instruction is generated and the executability is verified, if it does not match, it is re-adjusted; the feedback data is generated by analyzing the instruction execution deviation, and the scheduling algorithm parameters and key event extraction rules are updated.

[0105] It should be noted that, in this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0106] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and does not limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for pipeline detection for civil aviation security, characterized in that: Comprising the following steps: Step S101, real-time photographing of the aircraft refueling process through a high-resolution camera and image processing technology, and converting the image data into a digital format to form standardized video frames that can be analyzed; comprising: Capturing the aircraft refueling dynamic scene through a high-frame-rate camera to obtain a continuous image sequence; Compressing and denoising the continuous image sequence, and converting it into digital visual data when the definition condition is met; Storing, indexing, and extracting key frame features from the digital visual data to generate an analyzable visual data stream; Step S102, using a target detection algorithm to analyze the standardized video frame sequence, identifying personnel and equipment positions involved in the refueling process, and determining position coordinate data; comprising: Using a target detection algorithm to identify personnel and equipment positions and generate position coordinate data; Building a trajectory data set based on the position coordinate data, and extracting motion pattern features after smoothing processing; Combining a pre-set rule library to judge abnormal operations, labeling and generating a structured event data stream for the video frame sequence; The target detection algorithm performs real-time analysis based on the structured data generated in step S101, and the position coordinate data is pushed to the spatial analysis module of S103 and the real-time state database of step S107, providing spatial dimension basis for safety area judgment and dispatch optimization; Step S103, if the position coordinate data indicates that the personnel or equipment deviates from the pre-set safety area, analyze the standardized video frame sequence through a behavior classification model to determine whether there is an abnormal behavior, and obtain a behavior classification result; comprising: If the position coordinate data exceeds the boundary of the pre-set safety area, compare the position coordinate data with the safety area boundary through the spatial analysis module to determine the personnel position deviation or equipment position deviation; extract the video frame subset corresponding to the time period from the standardized video frame sequence according to the personnel position deviation or equipment position deviation, and obtain the deviation period frame sequence; analyze the deviation period frame sequence using a convolutional neural network to extract behavior pattern features and obtain a behavior pattern data set; if the behavior pattern data set does not match the pre-set behavior rule library, compare the behavior pattern features with the abnormal detection threshold to determine the abnormal behavior, and obtain an abnormal behavior identifier; label the corresponding frames from the video frame sequence according to the abnormal behavior identifier, and generate video frame data with abnormal labels; construct a time series index through the video frame data with abnormal labels, and generate a structured abnormal event data stream; use the structured abnormal event data stream to update the pre-set safety area boundary, and obtain optimized safety area data; Step S104, automatically identify whether the operation conforms to the standard process using a visual analysis algorithm, and mark risk event labels for operations that do not conform to the standard process; Step S105, sending the risk event labels and corresponding standardized video frame sequence segments to the monitoring center through data stream transmission technology to generate an alarm signal; Step S106, recording the alarm signal and standardized video frame sequence segment using all-weather data storage technology to generate a data record; Step S107, according to the data record and real-time operation state extraction technology, analyze the key events in the refueling process, adopt dynamic scheduling algorithm to adjust the operation process, optimize the refueling task scheduling, get the scheduling instruction.

2. A pipeline detection method for civil aviation security according to claim 1, characterized in that: In step S104, combined with the visual analysis algorithm, the operation whether conforms to the standard process is automatically identified, and the operation that does not conform to the standard process is marked with a risk event label, including: Process the video frame sequence through image segmentation technology, separate the refueling equipment and the operation personnel action; Extract the operation personnel action feature, generate the action mode vector and compare with the standard operation template database; Identify, label the non-compliant action, generate the structured non-compliant event data stream, and update the operation quality evaluation database.

3. A pipeline inspection method for civil aviation security according to claim 1, characterized in that: In step S105, the risk event label and the corresponding standardized video frame sequence segment are sent to the monitoring center through the data stream transmission technology to generate an alarm signal, including: Obtain the video frame segment marked with the risk event label, generate the standardized video frame data stream and send it to the monitoring center; When the number of risk event labels exceeds the preset threshold, generate an alarm signal data, extract a high-priority video segment; Process the high-priority video segment to generate structured event data, update the risk event database of the monitoring center and determine the dynamic alarm signal.

4. The method for pipeline inspection for civil aviation security according to claim 1, characterized in that: In step S106, the alarm signal and the standardized video frame sequence segment are recorded by using all-weather data storage technology to generate data records, including: Collect the alarm signal and the video frame sequence to form an original data set, and standardize the video frame sequence; If the frame rate is not up to standard, supplement the missing frames, align the time stamp of the complete frame sequence with the alarm signal; Analyze the synchronous data set to determine the event trigger point, extract the related data to generate data records and compress the storage.

5. A pipeline inspection method for civil aviation security according to claim 1, characterized in that: In step S107, according to the data record and real-time operation state extraction technology, analyze the key events in the refueling process, adopt dynamic scheduling algorithm to adjust the operation process, optimize the refueling task scheduling, get the scheduling instruction, including: Extract the key events from the data record and real-time state data and determine the priority; If there is a high-priority event, adjust the operation process by using dynamic programming algorithm to generate an optimized task scheduling scheme; Verify the validity of the scheduling parameters, generate the scheduling instruction and adjust the operation sequence in real time, and feed back the updated dynamic scheduling algorithm.

6. A pipeline inspection method for civil aviation security according to claim 5, characterized in that: Also includes: Extract the key events from the data record and the system state, determine the time stamp and the priority; When the priority of the key event is higher than the preset threshold, use the heuristic scheduling algorithm to sort the tasks and adjust the allocation scheme; Verify and update the task instruction combined with the real-time system state, generate the final scheduling instruction and verify its reliability.

7. A pipeline inspection method for civil aviation security according to claim 6, characterized in that: Also includes: Parse the data record to obtain the historical operation data, extract the key event set and determine the initial sequence of task allocation combined with the real-time state; If the initial sequence meets the scheduling constraint condition, optimize the task sequence, otherwise update the key event set; Generate the scheduling instruction and verify its executability, generate the feedback data according to the execution deviation, update the scheduling algorithm parameters and the key event extraction rules.

Images