Automatic identification method and device for key time nodes of airport ground guarantee work

Abstract

The invention discloses an airport ground guarantee work key time node automatic identification method and device, and relates to the technical field of image processing, and the method comprises the steps: inputting each airport scene image in an airport scene monitoring video into a trained target detection model, obtaining category information and position information of behavior subjects in each airport scene image; by using a target tracking model, according to category information and position information of key time node behavior subjects in all airport scene images, determining space-time motion information and confidence of each behavior subject in the airport scene monitoring video; and inputting the category information, the space-time motion information and the confidence coefficient of each behavior subject in the airport scene monitoring video into a trained dynamic graph attention model to obtain an identification result of the airport scene ground support work key time node. According to the invention, the accuracy and efficiency of important ground guarantee working time node identification in the airport scene image are improved.

Description

Technical Field

[0001] The present application relates to the technical field of image processing, and particularly to a method and device for automatically identifying key time nodes of airport ground support work. Background Art

[0002] In the airport surface operation support system, the management of key time nodes of aircraft ground support work is one of the core links of the Airport Collaborative Decision Making (A-CDM) system. The automatic acquisition of key time nodes refers to analyzing, identifying, and extracting the start / end times of each support work from video surveillance in real time, and quickly and accurately transmitting these time data to the A-CDM system to achieve information update and support the optimization of decision-making. The automatic identification and acquisition of time data for various ground support works in the airport apron area are of great significance for promoting the intelligent transformation of surface activity management and improving the departure efficiency.

[0003] However, at present, the acquisition of most airport ground support node times relies on manual methods, either by front-line support personnel manually recording or by supplementing the recording after notifying the duty room. This method not only has the problem of low efficiency, but is also easily affected by human omissions, resulting in data loss or errors. Therefore, an accurate and efficient automatic identification method is needed to monitor the start and end time nodes of surface support work. Summary of the Invention

[0004] The purpose of the present application is to provide a method and device for automatically identifying key time nodes of airport ground support work, which can accurately and efficiently identify important ground support work time nodes in airport surface images.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides a method for automatically identifying key time nodes of airport ground support work, including:

[0007] Obtain an airport scene monitoring video; the airport scene monitoring video includes a number of consecutive airport surface images;

[0008] For each of the airport surface images, input the airport surface image into a trained target detection model to obtain the category information and location information of the behavior subject in each airport surface image; the behavior subject is the equipment for implementing airport ground support work;

[0009] Using the target tracking model, based on the category information and location information of the actors at the key time nodes in all the airport scene images, determine the spatio-temporal motion information and confidence of each actor in the airport scene surveillance video; the spatio-temporal motion information includes the time information and location information of the actor in the airport scene surveillance video.

[0010] Input the category information, spatio-temporal motion information and confidence of each actor in the airport scene surveillance video into the trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work.

[0011] In a second aspect, the present application provides an automatic recognition device for key time nodes of airport ground support work, including:

[0012] An airport scene surveillance video acquisition module for acquiring an airport scene surveillance video; the airport scene surveillance video includes a number of consecutive airport scene images.

[0013] A target detection module for inputting each airport scene image into a trained target detection model for each airport scene image to obtain the category information and location information of the actors in each airport scene image; the actors are the equipment for implementing airport ground support work.

[0014] A target tracking module for using the target tracking model to determine the spatio-temporal motion information and confidence of each actor in the airport scene surveillance video based on the category information and location information of the actors at the key time nodes in all the airport scene images; the spatio-temporal motion information includes the time information and location information of the actor in the airport scene surveillance video.

[0015] A key time node recognition module for inputting the category information, spatio-temporal motion information and confidence of each actor in the airport scene surveillance video into the trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work.

[0016] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the above-mentioned automatic recognition method for key time nodes of airport ground support work.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned automatic recognition method for key time nodes of airport ground support work.

[0018] According to the specific embodiments provided by the present application, the present application discloses the following technical effects:

[0019] The present application provides a method and device for automatically identifying key time nodes in airport ground support work. By using a trained object detection model to detect the category information and position information of the acting subjects in airport scene images, and adopting a machine learning model to learn the rules of airport scene images and the category information and position information of the acting subjects in the images, the identification accuracy and efficiency of the acting subjects are improved; the spatio-temporal motion information and confidence of each acting subject in the airport scene monitoring video are obtained through an object tracking model, and the category information, spatio-temporal motion information and confidence of each acting subject are used as the input of a trained dynamic graph attention model. By using the dynamic graph attention model to learn the relationship between the dynamically changing spatio-temporal features (spatio-temporal motion information) of each acting subject, the accuracy and efficiency of automatically identifying key time nodes in airport ground support work are improved, and the problems of low efficiency and poor accuracy of traditional identification methods are solved. Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0021] Figure 1 It is an application environment diagram of a method for automatically identifying key time nodes in airport ground support work in Embodiment 1 of the present application;

[0022] Figure 2 It is a schematic flowchart of a method for automatically identifying key time nodes in airport ground support work provided in Embodiment 1 of the present application;

[0023] Figure 3 It is a schematic diagram of the specific process of a method for automatically identifying key time nodes in airport ground support work provided in Embodiment 1 of the present application;

[0024] Figure 4 It is a schematic diagram of key time nodes in typical aircraft ground support work during the aircraft turnover process provided in Embodiment 1 of the present application;

[0025] Figure 5 It is a schematic structural diagram of an improved YOLOv11 model provided in Embodiment 1 of the present application;

[0026] Figure 6 It is a flowchart of extracting target spatio-temporal information from airport scene images based on BoT-SORT provided in Embodiment 1 of the present application;

[0027] Figure 7 It is a flowchart of the BoT-SORT algorithm provided in Embodiment 1 of the present application;

[0028] Figure 8 Schematic diagram of the network architecture of the dynamic graph attention model provided in Embodiment 1 of the present application;

[0029] Figure 9 Overall system architecture solution based on ROS2 provided in Embodiment 1 of the present application;

[0030] Figure 10 Visualization interface for system deployment provided in Embodiment 1 of the present application;

[0031] Figure 11 Schematic diagram of the functional modules of an automatic recognition device for key time nodes of airport ground support work provided in Embodiment 2 of the present application;

[0032] Figure 12 Schematic diagram of the structure of a computer device provided in Embodiment 3 of the present application. Detailed implementation manners

[0033] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying 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. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0034] In order to effectively improve the management efficiency of airport surface activities, many existing surface monitoring technologies, such as surface movement radar (SMR), multilateration (MLAT), and automatic dependent surveillance-broadcast (ADS-B), have been widely used in target positioning and trajectory tracking. Although these technologies can provide the position information of the target, they only rely on the point signal characteristics of the surface target and lack the effective capture of the target behavior interaction and the external position information of the global form. Therefore, it is difficult to support the automatic acquisition of ground protection work node times in complex surface activities. Especially in the dynamic and complex surface environment of the airport, where there are many types of targets, complex behaviors, and affected by multiple factors such as weather, light changes, occlusion, and scale changes, the existing monitoring technologies have not been able to fully meet the real-time and accurate data acquisition requirements of airport surface ground support work time nodes. Based on the above problems, the present application provides an automatic recognition method and device for key time nodes of airport ground support work based on graph neural networks.

[0035] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0036] Embodiment 1

[0037] The method for automatically identifying key time nodes in airport ground support work provided by the embodiments of the present application can be applied to, for example, Figure 1 the application environment shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set separately, integrated on the server 104, placed in the cloud or on other servers. The terminal 102 can send the monitored video of the airport scene to be recognized to the server 104. After receiving the monitored video of the airport scene to be recognized, for the monitored video of the airport scene to be recognized, the server 104 inputs each airport scene image in the monitored video of the airport scene into the trained target detection model respectively to obtain the category information and position information of the behavior subject in each airport scene image, and uses the target tracking model to determine the spatio-temporal motion information and confidence of each behavior subject in the monitored video of the airport scene according to the category information and position information of the behavior subject at the key time nodes in all airport scene images. According to the category information, spatio-temporal motion information and confidence of each behavior subject in the monitored video of the airport scene, they are input into the trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work. The server 104 can feedback the obtained recognition result for the monitored video of the airport scene to the terminal 102. In addition, in some embodiments, the method for automatically identifying key time nodes in airport ground support work can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform recognition processing on the monitored video of the airport scene to be recognized, or the server 104 can obtain the monitored video of the airport scene to be recognized from the data storage system and perform recognition processing on the monitored video of the airport scene to be recognized.

[0038] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0039] In an exemplary embodiment, as Figure 2 and Figure 3 shown, a method for automatically identifying key time nodes in airport ground support work is provided. This method is executed by a computer device, and can specifically be executed independently by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to Figure 1 the server 104 in as an example for description, it includes the following steps 201 to step 204. Among them:

[0040] Step 201, obtain the airport scene monitoring video; the airport scene monitoring video includes several consecutive airport scene images.

[0041] Step 202, for each of the airport scene images, input the airport scene image into the trained object detection model to obtain the category information and location information of the acting body in each airport scene image; the acting body is the equipment for implementing airport ground support work.

[0042] Step 203, use the object tracking model to determine the spatio-temporal motion information and confidence of each acting body in the airport scene monitoring video according to the category information and location information of the acting body at the key time nodes in all the airport scene images; the spatio-temporal motion information includes the time information and location information of the acting body in the airport scene monitoring video.

[0043] Step 204, input the category information, spatio-temporal motion information and confidence of each acting body in the airport scene monitoring video into the trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work.

[0044] Implementing the above steps 201 to 204, detecting the category information and location information of the acting body in the airport scene image through the trained object detection model, and using the machine learning model to learn the rules between the airport scene image and the category information and location information of the acting body in the image, improving the recognition accuracy and efficiency of the acting body; obtaining the spatio-temporal motion information and confidence of each acting body in the airport scene monitoring video through the object tracking model, using the category information, spatio-temporal motion information and confidence of each acting body as the input of the trained dynamic graph attention model, and learning the relationship between the dynamically changing spatio-temporal features (spatio-temporal motion information) of each acting body through the dynamic graph attention model, improving the accuracy and efficiency of the automatic recognition of the key time nodes of the airport ground support work, and solving the problems of low efficiency and poor accuracy of the traditional recognition method.

[0045] This application first inputs the image data of the airport surface surveillance video into a trained object detection model based on the collected multi-source dataset of the airport surface to obtain the object detection results. Then, a Kalman filter and a multi-level matching strategy are used to obtain the spatio-temporal information of the behavior subjects in each frame of the video image. The output results are combined into a three-dimensional tensor for the airport surface surveillance video to be recognized and input into a dynamic graph attention model for processing, which is used for the classification of the key time nodes in the ground support work of the airport scene. It can complete the classification of 13 typical key time nodes in the ground support work of the airport and accurately record the start and end times of the time nodes. In addition, this application can also be implemented for engineering deployment based on the ROS2 platform to meet the requirements of the Civil Aviation Administration for the time error of node collection, that is, synchronize the camera video stream based on the ROS2 platform, integrate the detection, tracking, and classification modules to achieve real-time video stream processing, and the results can be synchronized to the Airport Collaborative Decision-making System (A-CDM) to realize the automatic recognition and classification of ground support work in complex apron scenarios. The key time node information of the aircraft ground support work is collected in real time through the surveillance video to improve the airport operation efficiency and collaborative decision-making ability. This application deeply combines object detection, multi-level data association, and spatio-temporal graph neural network in computer vision, significantly improving the automation level and decision-making efficiency of airport ground support work.

[0046] First, multi-source data from large hub airports, small and medium-sized airports, and simulation platform videos are collected to construct an annotated dataset with 13 categories of key time nodes including spatio-temporal positions and start and end time labels, denoted as the airport surface dataset. These 13 types of support work include: (1) Aircraft parking, (2) Aircraft departure, (3) Jet bridge docking, (4) Jet bridge evacuation, (5) Passenger staircase vehicle docking, (6) Passenger staircase vehicle separation, (7) Fuel truck refueling start, (8) Fuel truck refueling end, (9) Catering truck meal preparation start, (10) Catering truck meal preparation end, (11) Tractor docking, (12) Baggage loading and unloading start, (13) Baggage loading and unloading end. It covers eight types of target objects in the scene, namely aircraft, passenger staircase vehicles, catering trucks, fuel trucks, tractors, jet bridges, baggage trailers, and baggage chute vehicles. Figure 4 It is a schematic diagram of the key time nodes in the typical aircraft ground support work during the aircraft turnover process. The nodes in the dotted matrix are parallel processes, and the key time nodes shown in bold are the ground support work involved in this application. The method of this application is based on the constructed airport surface dataset for object detection, tracking, and key time nodes, which includes typical vehicles such as aircraft and passenger staircase vehicles, tractors, refueling vehicles, and catering trucks in the ground support work.

[0047] Before step 201, it is also necessary to train the object detection model, object tracking model, and dynamic graph attention model. The specific process is as follows:

[0048] 1. The training process of the object detection model includes the following steps A1 to A4.

[0049] Step A1, construct an object detection model; the object detection model in this embodiment is preferably the YOLOv11 model, and other object detection models based on neural networks can also be used. In this embodiment, the YOLOv11 model is improved, and the improved YOLOv11 model is used as the object detection model. In this improved YOLOv11 model, the MLLA (Mamba-Like Linear Attention) model is introduced into the attention module, and the Chain-of-Thought Prompted Adaptive Enhancer (CPA-Enhancer) is added before the first CBS module. In this embodiment, the improved one-stage object detection model YOLOv11 (i.e., the improved YOLOv11 model) is used, and its specific structure is as Figure 5 shown and applied to the detection process of typical vehicles and aircraft in ground support work. Since the YOLOv11 model belongs to the prior art, it will not be elaborated here. The specific improvement parts include:

[0050] (1) The MLLA model is introduced into the attention module, and some core designs of the Mamba model are incorporated into the linear attention mechanism. The Mamba model is centered on the State Space Model (SSM) and can capture the global dynamic relationship of the input data while avoiding the problem of over-reliance on global context in traditional methods. As Figure 5 shown, the MLLA model is added before each of the three detection heads in the detection head part. The detection head includes a first branch and a second branch, and both the first branch and the second branch include a CBS module, a CBS module, and a convolutional layer connected in sequence. The convolutional layer is a two-dimensional convolutional layer.

[0051] (2) The Chain-of-Thought network is introduced to dynamically generate a degradation scene enhancement strategy through the Chain-of-Thought prompting mechanism to cope with the inevitable states in daily life such as light changes, low visibility, foggy days, and rainy days. The RFAConv is used to extract the low-level features of the degraded image; the enhanced features are generated through four layers of encoders and decoders, and after being residually connected to the original image, they are input into the detection network; the CoT Prompt Generation Module (CGM) is used to dynamically learn the enhancement weights of degradation types such as foggy days and low light.

[0052] Step A2: Collect a number of visible light images of ground support vehicles and aircraft. In this embodiment, the acquisition sources of the dataset include large hub airports, small and medium-sized airports, and simulated platform videos. The videos include a number of visible light images. In this embodiment, Mixup and Mosaic target enhancement techniques are also used to linearly combine different visible light images to generate new training samples, which are denoted as visible light sample images. This method can effectively expand the training dataset, enabling the model to better adapt to different data distributions, reducing the risk of overfitting, and improving its robustness.

[0053] Step A3: Label the categories and positions of the objects in the above-mentioned number of visible light sample images respectively to obtain a number of labeled visible light sample images, and use the labeled visible light sample images as the labels of the visible light sample images. Among them, the objects refer to aircraft, jet bridges, passenger stairs vehicles, catering trucks, aviation fuel trucks, tractors, luggage trailers, and luggage slide trucks.

[0054] Step A4: Use a number of visible light sample images and the corresponding labeled visible light sample images to construct a training set and a test set, and train the object detection model to obtain a trained object detection model. The object detection model is trained using classification loss (Cls loss) and bounding box loss (Bbox loss).

[0055] Input the airport scene monitoring video into the trained object detection model to identify and locate the category information and location information of the behavior subjects related to the key time nodes frame by frame.

[0056] 2. The training process of the object tracking model includes: obtaining a first sample set; the first sample set includes a number of sample airport scene monitoring videos and the category information and location information of the behavior subjects at the key time nodes in each sample airport scene monitoring video; using the first sample set to train the object tracking model to obtain a trained object tracking model. In this embodiment, the object tracking model is the BoT-SORT model (BOT-SORT tracking algorithm).

[0057] Then step 202 specifically includes: combining the category information and location information of the actors related to the key time nodes in each frame of the airport scene monitoring video with the results predicted by the Kalman filtering method (Kalman filter), realizing the robust matching of the actors at the key time nodes of the ground support work, obtaining the spatial position changes of each actor, and obtaining the category information (identity label) and spatio-temporal motion information (temporal motion information) of the actor. Step 202 completes the extraction of continuous spatio-temporal information and trajectories based on object detection, uses the Kalman filtering method to predict the state of the actor to cope with the problems of uncertainty and noise in motion, and then determines the matching relationship between the predicted box and the detection box by combining the data association method of ReID and IoU information, ensuring the continuity and accuracy of the temporal motion information of the actors in the ground support work.

[0058] The process of extracting the spatio-temporal information of the objects in the airport scene image based on the BOT-SORT tracking algorithm and the BOT-SORT tracking algorithm process are respectively as Figure 6 and Figure 7 shown. Kalman filtering (Kalman) is applicable in both single-object and multi-object fields. In this application, Kalman filtering is regarded as a motion state model to predict the position of the actor (object), and the predicted result is used to correct the tracked object, so that Kalman filtering can be used to predict the position where the object appears in the subsequent frame of the airport scene image. In this embodiment, the Kalman filter is used to extract the spatio-temporal information of the actors in each frame of the airport scene image to obtain the prediction of the current frame of the airport scene image. In order to obtain the prediction result of the current frame of the airport scene image, the motion state model is initialized according to the single-frame detection result, and the initialized motion state model x is expressed as:

[0059]

[0060] where k represents the k-th frame of the airport scene image in the airport scene monitoring video sequence, (x c (k), y c (k)) is the center coordinate of the target bounding box recognized by the trained object detection model in the k-th frame of the airport scene image, x c (k) is the abscissa of the center point of the target bounding box in the k-th frame of the airport scene image, y c (k) is the ordinate of the center point of the target bounding box in the k-th frame of the airport scene image, w(k), h(k) are the width and height of the target bounding box in the k-th frame of the airport scene image respectively, are the velocity state quantities of the abscissa and ordinate of the center point of the target bounding box in the k-th frame of the airport scene image in the image coordinate system respectively, They are the velocity state variables of the width and height of the target bounding box in the k-th frame of the airport surface image in the image coordinate system, and T represents the transpose.

[0061] In this embodiment, the matrix forms of the state equation and the observation equation are as follows:

[0062] x k = Ax k-1 + Bu k + w k (2);

[0063] z k = Cx k + v k (3);

[0064] Among them, x k is the state vector at the k-th moment, A is the state transition matrix, representing the derivation of the state from the (k - 1)-th moment to the k-th moment, u k is the control input vector; B is the control input matrix, representing the relationship between the control input and the state; the process noise w k is a Gaussian white noise, conforming to w k ~ N(0, Q k ); Q k is the covariance matrix of the process noise; z k represents the observation vector (i.e., position) at the k-th moment; C represents the observation matrix, used to map the state vector to the observation space; the observation noise v k is also a Gaussian white noise, independent of the process noise w k , conforming to w k ~ N(0, R k ); R k The covariance is the covariance matrix of the observation noise.

[0065] In the prediction stage, the state prediction vector of the current k-th frame of the airport surface image is obtained based on the optimal state estimation of the (k - 1)-th frame of the airport surface image Obtained from the error estimation covariance matrix of the (k - 1)-th frame of the airport surface image and the measurement noise The prediction stage can be expressed as:

[0066]

[0067] In the formula, x k-1 is the state vector at the (k - 1)-th moment, is the predicted position estimate at the k-th moment, P k-1 is the position of the entity at the (k - 1)-th moment; Q k is the process noise covariance matrix.

[0068] In the correction stage, based on the state vector of the prediction stage and the measurement state to achieve the optimal estimation result. First, calculate the Kalman gain K k , then perform the optimal estimation of the state at time k, and finally update the error estimation covariance matrix at time k. The correction phase can be expressed as:

[0069]

[0070] In the formula, K k is the Kalman gain at time k, H k is the measurement matrix at time k, and P k is the position of the agent at the k-th moment.

[0071] The prediction and correction of the agent's position are carried out by continuously repeating the above process.

[0072] The data association between the prediction and the detection box adopts a data association method that combines ReID and IoU information. First, calculate the motion cost and appearance cost using the intersection over union (IoU) distance matrix and the cosine distance matrix respectively. For candidate boxes with low cosine similarity or large position distance, the algorithm will eliminate them. Finally, the minimum value of the two matrix elements is used as the element value of the cost matrix C. To facilitate data association, the intersection over union IoU between the prediction and the detection box is converted into a distance metric form, and the formula is as follows:

[0073]

[0074] This distance is used to measure the geometric similarity between the detection box and the prediction box. The smaller the value, the higher the matching possibility. The calculation formula of the cosine distance is as follows:

[0075]

[0076] Among them, represents the motion cost, which is the IoU distance between the i-th prediction box and the j-th detection box; represents the average appearance state of the i-th trajectory and the detection appearance state of the new target j cosine measurement value, represents the appearance cost; θ iou is a rejection threshold, which can be set to 0.5 to exclude trajectories with low possibility; θ emb is the ReID threshold, which is set to 0.25 to screen out the part that is positively correlated with the detection embedding vector from the appearance state of the trajectory. Combining the IoU distance and the cosine distance, the final cost matrix C is expressed as:

[0077]

[0078] Among them, C ijis the cost matrix for the element (i, j), that is, the cost matrix between the i-th predicted bounding box and the j-th detected bounding box. Figure 7 In the BOT-SORT tracking algorithm in Figure 7 , the airport scene monitoring video (i.e., a video frame sequence including several airport scene images) passes through a trained object detection model to obtain the detected bounding boxes and corresponding scores in each airport scene image. The detected bounding boxes are used to mark the positions of the behavioral subjects in the image. All detected bounding boxes are divided into high-score detected bounding boxes and low-score detected bounding boxes according to the scores corresponding to the detected bounding boxes. The airport scene monitoring video obtains the corresponding position estimation results for each airport scene image through the Kalman filtering algorithm. Based on the method combining ReID and IoU information, the high-score detected bounding boxes and the position estimation results are first associated and matched to obtain a first result. The first result includes the first matchable images and the first unmatchable images. Based on the IoU information method, the first unmatchable images are used as the remaining images to be secondarily associated and matched with the low-score detected bounding boxes to obtain a second result. The second result includes the images of the second matchable results and the second unmatchable images (the remaining images again).

[0079] The training process of the dynamic graph attention model includes: obtaining a second sample set; the second sample set includes the category information, spatio-temporal motion information, confidence of each behavioral subject in several sample airport scene monitoring videos, and the recognition results of the key time nodes of the airport scene ground support work in each sample airport scene monitoring video; using the second sample set to train the dynamic graph attention model to obtain a trained dynamic graph attention model. In this embodiment, the recognition results of the key time nodes of the airport scene ground support work include the time categories of the recognized key time nodes, as well as the category information and position information of the behavioral subjects in these key time nodes. In this implementation, step 204 above specifically includes the following steps B1 to step B3.

[0080] Step B1: Construct a three-dimensional tensor according to the category information, spatio-temporal motion information, and confidence of each behavioral subject in the airport scene monitoring video; the first dimension of the three-dimensional tensor corresponds to the time step in the airport scene monitoring video, the second dimension represents the number of behavioral subjects in each frame of the airport scene image, and the third dimension corresponds to the feature dimension of each behavioral subject; the feature dimension of the behavioral subject includes position information, category information, and confidence.

[0081] Obtain the airport surface surveillance video to be recognized, analyze each frame of the airport surface image in the video dataset frame by frame, extract the required information through the object detection and tracking model, and obtain key features such as the identity category and spatio-temporal location of the scene objects in each frame of the image. Merge the output results of the trained object tracking model into a three-dimensional tensor, and input it into the trained dynamic graph attention model for processing, which is used for the classification of critical times in the ground support work of the airport scene. It can complete the classification of 13 typical critical time nodes in the ground support work of the airport scene, and accurately record the start and end times of the critical time nodes.

[0082] Step B2: Construct a spatio-temporal dynamic graph based on the three-dimensional tensor; the nodes of the spatio-temporal dynamic graph correspond to the behavior subjects; the edges of the spatio-temporal dynamic graph are obtained by connecting the nodes corresponding to the behavior subjects.

[0083] After obtaining the key features such as the identity category and spatio-temporal location of the scene objects in each frame of the airport surface image, the information generated by each video data containing T frames of airport surface images can be represented as a three-dimensional tensor and transformed into graph-structured information. Analyze each frame of the airport surface image in the video dataset frame by frame, and extract and obtain the key features such as the category and spatio-temporal location of the behavior subjects in each frame of the airport surface image through the object detection model and the object tracking model. The information generated by each video data containing T frames of airport surface images can be represented as a three-dimensional tensor. The first dimension of this tensor corresponds to the time step in the video, the second dimension represents the number of objects (i.e., the number of nodes) in each frame of the airport surface image, and the third dimension is the feature dimension of each object, including attributes such as the spatial coordinates, category, and confidence of the object. Therefore, when given an airport surface surveillance video containing T frames of airport surface images, this embodiment proposes a dynamic graph construction method to construct a fully connected dynamic graph G=(V, E). Specifically, the node set is represented as: V={v (t,i) |t = 1, …, N}, where v (t,i) is the i-th behavior subject in the airport surface image corresponding to the t-th time step of the airport surface surveillance video, and this node set represents all the behavior subjects in the airport surface surveillance video sequence. Determine the node features and edge features of the spatio-temporal dynamic graph. The node features include the feature dimensions of all behavior subjects in the airport scene surveillance video, and the node feature vector is represented as: F={f (t,i) |t = 1, …, N}, where f (t,i) represents the feature vector of node v (t,i) , which is represented by the x coordinate and y coordinate of the center point of the target bounding box, the width and height of the target bounding box, and the category. The edge set E of the spatio-temporal dynamic graph can include three subsets: a spatial edge subset, a temporal edge subset, and a self-connection edge subset, which further enhances the model's ability to model spatio-temporal relationships: E={E se , E te , Esce}, where the spatial edge subset characterizes the connection relationship between different nodes within the same time step, expressed as: E se = v (t,i) → v (t,j) (i ≠ j), v (t,j) is the j-th agent in the airport surface image corresponding to the t-th time step of the airport surface surveillance video. The temporal edge subset characterizes the connection relationship between nodes in different time steps, expressed as: E te = v (t,i) → v (k,j) (t ≠ k), v (k,j) is the j-th agent in the airport surface image corresponding to the k-th time step of the airport surface surveillance video. The self-connection edge subset characterizes the connection relationship of a node with itself, expressed as: E sce = v (t,i) → v (t,i) .

[0084] Step B3: Input the node features and edge features of the spatio-temporal dynamic graph into the trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work. The network model architecture of the trained dynamic graph attention model is as Figure 8 shown. The dynamic graph attention model includes a sequentially connected sequential processing stack, a spatial attention module ( Figure 8 the spatial attention layer in Figure 8 ), a temporal attention module (

[0085] the temporal attention layer in ), a global average pooling layer (Average Pooling), a fully connected layer (FCLayer), and a SoftMax layer; the sequential processing stack includes a linear layer, a fully connected layer, a ReLU activation function, and a layer normalization layer.The traditional graph attention network is extended to the spatio-temporal domain through a spatio-temporal graph attention network (i.e., a dynamic graph attention model) to effectively process complex dynamic graph data with node topologies and attributes changing dynamically over time. A fixed-time interval sliding window method based on the aircraft movement trajectory is combined for temporal positioning to implement aircraft ground support work based on graph neural networks. The dynamic graph attention model comprehensively extracts spatio-temporal features using dynamic modeling methods to identify and classify key time nodes. Its processing flow can be divided into the following steps: First, the input data (the input data is temporal data including the category information, spatio-temporal movement information, and confidence of each actor in the airport scene monitoring video) passes through a sequential processing stack, which includes modules such as a linear layer (Reshape), a fully connected layer, a ReLU activation function, layer normalization (LayerNormalization, LN), and a Dropout layer. The input data size of the sequential processing stack can be S×T×N×F, and it outputs preprocessed node features of size S×(T×N)×128. The design of this sequential processing stack aims to introduce non-linear transformation and normalize the node features, thereby enhancing the model's expressive ability and stability to obtain preprocessed node features. After the preprocessing of the sequential processing stack, the preprocessed node feature data is sequentially input into the spatial attention module and the temporal attention module to extract the features of the actors (scene objects) in space and time respectively, obtaining spatio-temporal attention features. Specifically, the preprocessed node features are input into the spatial attention module to obtain spatial attention features. The spatial attention features are input into the temporal attention module to obtain spatio-temporal attention features. The spatial attention module focuses on the association between nodes at the same time step, while the temporal attention module focuses on the temporal correlation between adjacent time steps, thus ensuring that the model can accurately capture the node relationships in dynamic changes. After the feature extraction is completed, the spatio-temporal attention features are processed through a global average pooling layer to compress multi-dimensional feature information and generate a global descriptive vector, denoted as the pooled feature vector. Next, the pooled feature vector is flattened through a fully connected layer to obtain the flattened features, and finally, the flattened features are recognized through a SoftMax layer to obtain the recognition results of the key time nodes of the airport scene ground support work, realizing multi-class classification of the key time nodes, thus completing automatic recognition.

[0086] Figure 8The dynamic graph attention model shown in has two main modules, namely the spatial attention module and the temporal attention module, which aim to capture the changes in node features in the spatio-temporal dimension of the dynamic graph structure to more accurately model the time-varying node relationships. The spatial attention module includes a Spatial Positional Encoding layer, a Muti HeadedAttention layer, and a final sequential processing layer (ft_map). The temporal attention module includes a Temporal Positional Encoding layer, a MutiHeadedAttention layer, and a final sequential processing layer (ft_map). The sequential processing layer (ft_map) includes a fully connected layer (FCLayer), a ReLU activation function, a Layer Normalization (LN), and a Dropout layer connected in sequence. The positional encoding layer is used to identify the temporal order of key-frame airport scene images and the spatial position information of each agent, enabling the model to accurately understand and utilize the spatio-temporal correlations between different frames. In the multi-head attention layer, the input data is decomposed and combined at a fine-grained level by multiple attention heads of the model to fully capture the multi-level relationships between nodes. For the input feature representation f of node v in the constructed dynamic spatio-temporal graph (t,i) of (t,i) . Specifically, assuming that in the constructed spatio-temporal dynamic graph, the input feature of node v (t,i) is f (t,i) , for the h-th spatial attention head, the model first uses three fully connected layers to map f (t,i) to key, query, and value vectors. This process can be expressed as:

[0087]

[0088] where and represent the key vector, query vector, and value vector of the node respectively; and are the weight matrices of the key vector, query vector, and value vector of the node respectively.

[0089] The spatial attention layer A S is responsible for analyzing the spatial dependence relationships between nodes and evaluating the correlations of each node within the same time step. The spatial attention head calculates the weights of the spatial edges and self-connection edges in two steps. First, it calculates the scaled dot product of the query vector and the key vector of each node in the current time step, and then normalizes it through the SoftMax function to generate the attention weights. The formula is as follows:

[0090]

[0091] Among them, d is the dimension of the three vectors of the key vector, query vector, and value vector (assumed to be 32). Represents node v (t,i) and v (t,j) The scaled dot product between them; <.,.> represents the inner product operation. Is the attention weight between node v (t,i) and v (t,j) to measure the information transfer weight from node v (t,i) to node v (t,j) In addition, to maintain the independence between time steps, the network sets the weight of the time edge to 0 to avoid information transfer across time steps.

[0092] The spatial attention feature is calculated by representing the spatial feature of the node as the weighted sum of all value vectors within the current time step. The formula is as follows:

[0093]

[0094] Among them, Is the spatial attention feature of node v (t,i) The basic principle is that in the same time step, the node receives the information of other nodes according to the learned attention weight and aggregates it to form a feature representation based on the spatial relationship.

[0095] The final output of the spatial attention module is to concatenate the features extracted by all spatial attention heads, thereby obtaining the complete spatial feature representation of node v (t,i) It can be specifically represented by the following formula:

[0096]

[0097] Among them, Concat represents the concatenation operation, H is the number of spatial attention heads in the spatial attention module (assumed to be 8). Through the multi-head attention mechanism, the model can understand the spatial structure characteristics between nodes from different perspectives. Is the spatial attention feature, Encodes various structural information learned from different attention heads, thereby adapting to the dynamics of node relationships in the spatio-temporal dynamic graph.

[0098] The key time node automatic recognition method for airport ground support work provided in this embodiment further includes: constructing a distributed system based on the ROS2 platform, integrating a detection module, a tracking module, and a classification module to implement real-time processing of airport scene monitoring video streams and visualization of the recognition results of key time nodes for airport scene ground support work; the detection module is used to input the airport surface image into a trained target detection model to obtain the category information and location information of the behavior subjects in each airport surface image; the tracking module is used to utilize a target tracking model to determine the spatio-temporal motion information and confidence of each behavior subject in the airport scene monitoring video according to the category information and location information of the behavior subjects at key time nodes in all airport surface images; the classification module is used to input the category information, spatio-temporal motion information, and confidence of each behavior subject in the airport scene monitoring video into a trained dynamic graph attention model to obtain the recognition results of key time nodes for airport scene ground support work.

[0099] The distributed system includes: a multi-view monitoring camera group that supports RTSP protocol video stream acquisition; an edge computing unit equipped with an improved YOLOv11, BoT-SORT, and spatio-temporal graph attention model to complete real-time inference; a visualization terminal that displays target trajectories and node logs. As Figure 9 shown, the processing process of the distributed system constructed based on the ROS2 platform specifically includes: the distributed system takes the RTSP video stream of the surface monitoring as the input. First, data preprocessing is performed through an rtsp client, including using the Python OpenCV library to read the stream and convert the data format, and then publishing it to the single-frame image spatial information detection and video image temporal information tracking nodes in the form of a topic. These nodes complete the acquisition of the spatio-temporal information of the surface behavior subjects and divide the results into three types of topics: ① video result drawing, ② single-frame image detection log, and ③ video temporal information. The first topic draws a tracking box, category information, and confidence on the image and transmits it to the airport scene monitoring video result drawing node; the second topic displays the currently recognizable behavior subjects (aircraft and surface special vehicles) in the interface log box; the third topic transmits the temporal information of the current behavior subjects to the next time node classification model. The airport scene monitoring video result is then transmitted to the video result drawing topic, and the single-frame image detection log is sent to the display log node. After the temporal information is processed by the nodes, it is published to the time node log topic and further transmitted to the log display node. Finally, the interface display node integrates these data, synchronously presents the airport surface image and text information on the interface, and communicates with the RTSP server through the control image capture topic to maintain the real-time performance of the system. Through the cyclic framework design, the distributed system realizes real-time processing and continuous operation. The visualization interface of the distributed system is as Figure 10As shown, there are camera 1, camera 2, camera 3 and lidar. For example, the identified key node logs include the key time node of the aircraft docking at 11:15:21 on October 19, 2024 and the key time node of the aircraft docking at 11:15:47 on October 19, 2024, and the object detection logs include the aircraft at 11:15:09 on October 19, 2024 and the aircraft at 11:15:39 on October 19, 2024.

[0100] This application also provides an application scenario that applies the above-mentioned automatic recognition method for key time nodes in airport ground support work. Specifically: The automatic recognition method for key time nodes in airport ground support work provided in this embodiment can be applied in the airport ground support scenario. The airport ground support scenario includes a video acquisition link and a key time node recognition link; the airport scenario monitoring video enters the key time node recognition link from the video acquisition link, obtains the corresponding recognition result through a human-machine collaboration method, and enters the downstream content distribution link. The automatic recognition method for key time nodes in airport ground support work provided in this embodiment belongs to the machine marking link in the key time node recognition link. Specifically, in the process of the key time node recognition link for the airport scenario monitoring video, each airport scene image in the airport scenario monitoring video can be input into the trained object detection model respectively to obtain the category information and position information of the behavior subject in each airport scene image, and the target tracking model is used to determine the spatio-temporal motion information and confidence of each behavior subject in the airport scenario monitoring video according to the category information and position information of the behavior subject at the key time node in all airport scene images. According to the category information, spatio-temporal motion information and confidence of each behavior subject in the airport scenario monitoring video, they are input into the trained dynamic graph attention model to obtain the recognition result of the key time node of the airport ground support work.

[0101] Embodiment 2

[0102] Based on the same inventive concept, this embodiment also provides an automatic recognition device for key time nodes in airport ground support work for implementing the above-mentioned automatic recognition method for key time nodes in airport ground support work. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the automatic recognition device for key time nodes in airport ground support work provided below can refer to the limitations on the automatic recognition method for key time nodes in airport ground support work in the above text, and will not be repeated here.

[0103] In an exemplary embodiment, as Figure 11 shown, an automatic recognition device for key time nodes in airport ground support work is provided, which includes the following modules.

[0104] The airport scene monitoring video acquisition module T1 is used to acquire the airport scene monitoring video; the airport scene monitoring video includes a number of consecutive airport scene images.

[0105] The target detection module T2 is used to input each airport scene image into a trained target detection model for each airport scene image to obtain the category information and location information of the acting subject in each airport scene image; the acting subject is the equipment for implementing airport ground support work.

[0106] The target tracking module T3 is used to use a target tracking model to determine the spatio-temporal motion information and confidence of each acting subject in the airport scene monitoring video according to the category information, location information of the acting subject at key time nodes in all the airport scene images; the spatio-temporal motion information includes the time information and location information of the acting subject in the airport scene monitoring video.

[0107] The key time node recognition module T4 is used to input the category information, spatio-temporal motion information and confidence of each acting subject in the airport scene monitoring video into a trained dynamic graph attention model to obtain the recognition result of the key time nodes of the airport scene ground support work.

[0108] Embodiment 3

[0109] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 12 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store key time node recognition data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for automatically recognizing key time nodes of airport ground support work.

[0110] Those skilled in the art can understand, Figure 12The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0111] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0112] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0114] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0115] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0116] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0117] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for automatically identifying key time nodes of airport ground support work, characterized in that: The method for automatically identifying key time nodes of airport ground support work includes: Acquire an airport scene surveillance video; the airport scene surveillance video includes a plurality of frames of continuous airport scene images; For each of the airport scene images, the airport scene image is input into a trained target detection model to obtain category information and location information of the subject in each of the airport scene images; the subject is the equipment that implements the airport ground support work; Using the target tracking model, according to the category information and position information of the subject at the key time nodes in all the airport scene images, the spatiotemporal motion information and confidence of each subject in the airport scene monitoring video are determined; the spatiotemporal motion information includes the time information and position information of the subject in the airport scene monitoring video; According to the category information, spatiotemporal motion information and confidence of each behavior subject in the airport scene surveillance video, they are input into the trained dynamic graph attention model to obtain the recognition results of the key time nodes of the ground support work in the airport scene.

2. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: The target detection model is an improved YOLOv11 model; the improved YOLOv11 model introduces the MLLA model in the attention module and adds the chain thinking network before the first CBS module.

3. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: The target tracking model is a BoT-SORT model.

4. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: According to the category information, spatiotemporal motion information and confidence of each subject in the airport scene surveillance video, the trained dynamic graph attention model is input to obtain the recognition results of the key time nodes of the airport scene ground support work, including: A three-dimensional tensor is constructed according to the category information, spatiotemporal motion information and confidence of each behavior subject in the airport scene monitoring video; the first dimension of the three-dimensional tensor corresponds to the time step in the airport scene monitoring video, the second dimension represents the number of behavior subjects in each frame of the airport scene image, and the third dimension corresponds to the feature dimension of each behavior subject; the feature dimension of the behavior subject includes position information, category information and confidence; Constructing a spatiotemporal dynamic graph based on a three-dimensional tensor; the nodes of the spatiotemporal dynamic graph correspond to behavioral subjects; the edges of the spatiotemporal dynamic graph are obtained by connecting the nodes corresponding to the behavioral subjects; The node features and edge features of the spatiotemporal dynamic graph are input into the trained dynamic graph attention model to obtain the recognition results of key time nodes of ground support work in airport scenes; the node features include the feature dimensions of all behavioral subjects in the airport scene monitoring video; the edge features include spatial edge subsets, temporal edge subsets and self-connected edge subsets; the spatial edge subset represents the connection relationship between different nodes in the same time step; the temporal edge subset represents the connection relationship between nodes in different time steps; the self-connected edge subset represents the connection relationship between a node and itself.

5. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: The dynamic graph attention model includes a sequential processing stack, a spatial attention layer, a temporal attention module, a global average pooling layer, a fully connected layer and a SoftMax layer connected in sequence; the sequential processing stack includes a linear layer, a fully connected layer, a ReLU activation function and a layer normalization layer.

6. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: The method for automatically identifying key time nodes of airport ground support work also includes: Build a distributed system based on the ROS2 platform, integrate detection modules, tracking modules and classification modules, realize real-time airport scene monitoring video stream processing and visualization of recognition results of key time nodes of airport scene ground support work; The detection module is used to input the airport scene image into the trained target detection model to obtain the category information and location information of the behavior subject in each airport scene image; The tracking module is used to determine the spatiotemporal motion information and confidence of each behavior subject in the airport scene monitoring video by using the target tracking model according to the category information and position information of the behavior subject at key time nodes in all the airport scene images; The classification module is used to input the category information, spatiotemporal motion information and confidence of each behavior subject in the airport scene monitoring video into the trained dynamic graph attention model to obtain the recognition results of the key time nodes of the airport scene ground support work.

7. The method for automatically identifying key time nodes of airport ground support work according to claim 1 is characterized in that: The categories of the actors include aircraft, passenger elevators, catering trucks, aviation fuel trucks, tractors, jet bridges, luggage trolleys and luggage slides; the key time nodes include aircraft positioning, aircraft departure, jet bridge docking, jet bridge evacuation, passenger elevator docking, passenger elevator separation, start of aviation fuel truck refueling, end of aviation fuel truck refueling, start of catering truck catering, end of catering truck catering, tractor docking, start of luggage loading and unloading, and end of luggage loading and unloading.

8. An automatic identification device for key time nodes of airport ground support work, characterized in that: The automatic identification device for key time nodes of airport ground support work includes: An airport scene monitoring video acquisition module is used to acquire an airport scene monitoring video; the airport scene monitoring video includes a plurality of frames of continuous airport scene images; The target detection module is used for inputting each of the airport scene images into a trained target detection model to obtain the category information and location information of the subject in each of the airport scene images; the subject is the equipment that implements the airport ground support work; A target tracking module is used to determine the spatiotemporal motion information and confidence of each behavior subject in the airport scene monitoring video by using a target tracking model according to the category information and position information of the behavior subject at key time nodes in all the airport scene images; the spatiotemporal motion information includes the time information and position information of the behavior subject in the airport scene monitoring video; The key time node recognition module is used to input the category information, spatiotemporal motion information and confidence of each behavior subject in the airport scene monitoring video into the trained dynamic graph attention model to obtain the recognition results of the key time nodes of the airport scene ground support work.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for automatically identifying key time nodes of airport ground support work as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for automatically identifying key time nodes of airport ground support work described in any one of claims 1 to 7 is implemented.

Images