A sample database construction method for an aerial refueling scene

Abstract

The application discloses a kind of sample database construction methods for aerial refueling scene, it is related to database field, comprising: setting extreme, conventional and special meteorological environment category, based on each category corresponding aerial refueling process video and image data are collected;Screening collected video, image data, eliminate picture blur, key target missing and the data of shooting error exceeding standard;The application is by subdividing meteorological environment category and accurately calculating influence coefficient, ensure that data covers comprehensively and is close to actual aerial refueling scene, when collecting data, multi-dimensional screening is removed inferior data, guarantee data quality, according to refueling stage setting different time interval extraction image frame and synchronous accurate time stamp, so that data is more time effectiveness and pertinence.

Description

Technical Field

[0001] This invention relates to the field of database technology, specifically to a method for constructing a sample database for an aerial refueling scenario. Background Technology

[0002] The invention patent application with publication number CN109085845B discloses an autonomous aerial refueling docking bionic visual navigation control system and method. Its purpose is to provide a navigation control system and method in the process of soft autonomous aerial refueling docking, aiming to improve the reliability, anti-interference and accuracy of close-range relative navigation in the autonomous aerial refueling docking stage, and to design a matching relative position precision control method with control switching, thereby improving the accuracy of close-range navigation and control, promoting the smooth realization of soft autonomous aerial refueling, and improving the autonomous capability level of UAVs.

[0003] However, to meet the refueling docking requirements in aerial refueling missions, it is necessary to collect video, image and other data under different weather conditions to build an aerial refueling sample database, thereby providing data support for the training or research of aerial refueling related algorithms.

[0004] To address this, we propose a method for constructing a sample database for aerial refueling scenarios. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a sample database construction method for aerial refueling scenarios, which can effectively solve the problems of the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions;

[0007] This invention discloses a method for constructing a sample database for aerial refueling scenarios, comprising:

[0008] Meteorological parameters are acquired, and the dynamic coupling influence coefficient of these parameters on aerial refueling safety is calculated. The meteorological environment category is determined based on this coefficient: an extreme meteorological environment category is defined as a coefficient ≥ 8, a normal meteorological environment category as a coefficient ≤ 3, and a special meteorological environment category as a coefficient < 3 < 8. Video and image data of the aerial refueling process are collected for each category. The collected video and image data are then filtered, removing blurry images, missing key targets, and images with excessive shooting errors. Image frames from the filtered videos are extracted at fixed time intervals, and the image frames and original image data are converted to a preset standard format. Meteorological environment parameters, the location of key aerial refueling components, and refueling stage annotations are added to the converted data. Simultaneously, a feature extraction and localization model for key aerial refueling components is constructed, comprising a component feature enhancement module, a multi-scale feature matching module, and a location coordinate optimization module. The module performs grayscale layering on the converted image frames and the original image data, filters key feature intervals, and enhances key component features through linear stretching and compression of grayscale values. The multi-scale feature matching module pre-sets a standard feature template library for the refueling probe of the refueling aircraft and the refueling port of the receiving aircraft, calculates the similarity between the key component features and each standard feature template in the standard feature module library, and selects the initial positioning area of ​​the key component based on the similarity. The position coordinate optimization module uses the initial positioning area as a reference, scans the area boundary using the 8-neighbor pixel traversal method, determines the minimum bounding rectangle of the key component, and calculates the relative positional relationship parameters of the key component. The module constructs a database infrastructure, classifies and stores labeled data according to meteorological environment categories and refueling stages, and simultaneously establishes associated indexes. The module verifies the amount of video and image data and the completeness of labeled entries under each meteorological environment category in the database, supplements aerial refueling video and image data for categories with insufficient data, and corrects meteorological parameters, component positions, and stage division content in the labeled information that do not match the actual scene.

[0009] Furthermore, the dynamic coupling influence coefficient is calculated using the following formula:

[0010] ;

[0011] Among them, the single-factor nonlinear response function is:

[0012] ;

[0013] Two-factor coupling function:

[0014] ;

[0015] Time-domain rate of change function:

[0016] ;

[0017] Relative position safety margin correction function:

[0018] ;

[0019] In the formula: This represents the dynamic coupling influence coefficient. These correspond to four meteorological factors: wind speed, precipitation, temperature difference, and visibility. The basic weight coefficient for the p-th meteorological factor; For the p-th and q-th meteorological factors, the coupling weight coefficient is denoted as . This is the weighting coefficient for the rate of change in the time domain; This refers to the relative position safety margin weighting coefficient; Let p be the measured value of the p-th meteorological factor; The safety threshold for the p-th meteorological factor; This is the reference standard value for the p-th meteorological factor; , Let be the nonlinear response parameter of the p-th meteorological factor; , Let be the coupling strength coefficient between the p-th and q-th meteorological factors; , , Let be the time-domain dynamic parameter of the p-th factor; These are the relative position parameters of the two machines; The optimal relative position parameters; For the current safety margin; This represents the critical safety margin. The standard deviation of the relative position tolerance; Adjust the weighting coefficients for the location.

[0020] Furthermore, the filtering and removal operations for the collected video and image data include:

[0021] Calculate the resolution evaluation index of video or image. , Indicates the height and width in pixels of a video frame or image. This represents the gradient value of pixel (i,j) in the x-direction. This represents the gradient value of pixel (i,j) in the y-direction. < At that time, the image was determined to be blurry and was removed. Set a preset resolution threshold;

[0022] A pre-defined set of key targets for aerial refueling is used, including the refueling probe of the tanker aircraft, the refueling port of the receiver aircraft, the fuselage markings of the tanker aircraft and the receiver aircraft. Target detection algorithms are used to identify key targets in videos or images. If any key target is missing in the identification results, it is determined to be a missing key target and is removed.

[0023] Calculate the shooting angle deviation value And the shooting distance deviation value D, when > And D > If the shooting error exceeds the standard, it will be rejected.

[0024] in, To preset the viewing angle deviation threshold, To preset the distance deviation threshold, The angle between the actual shooting angle and the standard shooting angle is represented by , and D represents the difference between the actual shooting distance and the standard shooting distance.

[0025] Furthermore, the operation of extracting image frames from the filtered video at fixed time intervals includes:

[0026] The refueling process is divided into four phases: preparation and docking, precise docking, stable refueling, and separation and withdrawal. Fixed time intervals are set for each phase.

[0027] When extracting image frames, the timestamp corresponding to each image frame is recorded synchronously, with the timestamp accurate to the millisecond level, and it is synchronized with the real-time clock system of the aerial refueling scenario.

[0028] Furthermore, when converting image frames and raw image data into a preset standard format, the preset standard format parameters include image resolution, color space, and image compression format;

[0029] During the conversion process, interpolation algorithms are used to adjust the resolution of non-standard resolution image frames and original image data, and a color space conversion matrix adapted to the aerial refueling scenario is used to convert the non-RGB color space image data to color space.

[0030] ;

[0031] In the formula: R, G, and B are the red, green, and blue component values ​​of the RGB color space, respectively; Y, U, and V are the luminance, blue difference, and red difference component values ​​of the YUV color space, respectively.

[0032] After the conversion is completed, the image data is checked for integrity. If the check finds that the image data is damaged or missing, the conversion operation is repeated.

[0033] Furthermore, when adding meteorological environmental parameter annotation information to the format-converted data, real-time meteorological environmental parameters are obtained by sensing through an air-ground cooperative sensor network installed on the refueling aircraft and the receiving aircraft. The air-ground cooperative sensor network includes wind speed sensors, precipitation sensors, temperature sensors, and visibility sensors.

[0034] The collected real-time meteorological and environmental parameters are preprocessed synchronously, including outlier removal and adaptive smoothing. The outlier removal operation is as follows:

[0035] When any collected data satisfy When an outlier is detected, it is identified and removed. This is the mean of the data collected for this parameter. The standard deviation of the data collected for this parameter;

[0036] The adaptive smoothing process is as follows:

[0037] ;

[0038] In the formula: The value of the meteorological parameter at time t is the smoothed value. To adapt to window size; As a time decay weight; represents the raw values ​​of meteorological parameters collected at time k; a and b are adjustment coefficients;

[0039] The preprocessed meteorological environmental parameters are matched with the format-converted data through time correlation, so that each image frame and the original image data correspond to a unique set of meteorological environmental parameter annotation information.

[0040] Furthermore, when adding location annotations for key aerial refueling components to the converted data, a feature extraction and localization model for these components is simultaneously constructed. This model includes a component feature enhancement module, a multi-scale feature matching module, and a location coordinate optimization module.

[0041] The component feature enhancement module performs grayscale layering processing on the format-converted image frames and the original image data, and selects grayscale intervals with texture density greater than or equal to 0.6 and edge gradient ratio greater than or equal to 0.5 as key feature intervals. The key component features are enhanced by linear stretching and compression of grayscale values.

[0042] The multi-scale feature matching module uses a pre-defined feature template library of refueling probes and receiving ports of fuel dispensers to calculate feature similarity.

[0043] S = ∑ m = 1 M ∑ n = 1 N [ F ( m , n ) × T ( m , n ) ] ∑ m = 1 M ∑ n = 1 N F ( m , n ) 2 × ∑ m = 1 M ∑ n = 1 N T ( m , n ) 2 ;

[0044] In the formula: These are the height and width pixels of the feature template, respectively. It is the feature value at position (m,n) within the key feature interval of the image, that is, the gray value at that position after gray-scale stretching; For the feature value at position (m,n) in the standard feature template, i.e., the standard gray value at the corresponding position in the template, select the similarity. The image area corresponding to the standard template with a value ≥0.85 is used as the initial positioning area for key components;

[0045] The position coordinate optimization module uses the initial positioning area as a reference, employs an 8-neighbor pixel traversal method to scan the area boundary, determines the minimum bounding rectangle of key components, and calculates the relative positional parameters of key components, including the straight-line distance between the refueling probe of the fuel dispenser and the receiving port of the receiving unit. and angle deviation The calculation formula is:

[0046] ;

[0047] In the formula: ( ( ) represents the center coordinates of the rectangular bounding box of the fuel dispenser's fuel probe; () represents the center coordinates of the rectangular bounding box of the receiver's oil inlet; Standard angles for aligning key components during aerial refueling;

[0048] The location coordinates and relative positional parameters of key components are added to the corresponding data as location annotation information.

[0049] Furthermore, when adding refueling stage annotation information to the format-converted data, the refueling stage determination model is constructed through the following steps:

[0050] The core feature parameters of the aerial refueling process are selected as the model input variables. These core feature parameters include the real-time relative distance d between the tanker and the receiver aircraft, the real-time relative speed v, and the alignment accuracy of key components. ;

[0051] Standardize the selected feature parameters;

[0052] The determination logic is set as follows:

[0053] When the standardized relative distance exceeds 0.8 and the standardized relative speed exceeds 0.7, it is determined to be in the docking preparation stage;

[0054] When the standardized relative distance is in the range of [0.1, 0.8], the standardized relative speed is in the range of [0.2, 0.7], and the alignment accuracy of key components is less than 90%, it is judged to be in the precise docking stage;

[0055] When the standardized relative distance is less than 0.1, the standardized relative speed is less than 0.2, and the alignment accuracy of key components is not less than 90%, it is determined to be in the stable refueling stage.

[0056] When the standardized relative distance exceeds 0.3, the standardized relative speed exceeds 0.5, and the alignment accuracy of key components is <50%, it is determined to be the separation and withdrawal stage;

[0057] The constructed judgment model is validated using a pre-set historical aerial refueling annotation dataset. If the accuracy is lower than 98%, the judgment threshold range for each stage is adjusted.

[0058] Furthermore, the database infrastructure includes a data storage layer, an index management layer, and an access interface layer:

[0059] The data storage layer adopts a segmented storage method, which is divided into extreme weather storage segments, normal weather storage segments, and special weather storage segments according to meteorological environment categories. Each meteorological storage segment is further divided into sub-storage areas based on the refueling stage.

[0060] The index management layer is set up with multi-level related indexes: the first-level index uses the meteorological environment category as the index key, the second-level index uses the refueling stage as the index key, and the third-level index uses the data collection timestamp as the index key. The index structure adopts a B+ tree index structure.

[0061] The access interface layer is configured to support both SQL queries and API calls for data access.

[0062] Furthermore, the verification, completion, and correction operations in the database are as follows:

[0063] When verifying the data volume, a minimum threshold for the video and image data volume under each meteorological environment category is preset. If the data volume under a certain meteorological environment category is lower than the corresponding minimum threshold, the data completion process is initiated.

[0064] When verifying the completeness of labeled entries, calculate the label completeness evaluation index:

[0065] , To label the integrity evaluation indicators, To determine the number of entries with accurate and complete information. The total number of all labeled entries in the database, when If the percentage is less than 95%, the item is deemed to be incomplete and the incomplete item needs to be manually completed.

[0066] When correcting the annotation information, the meteorological parameters, component locations, and phase divisions in the annotation information are compared with the real-time recorded data of the aerial refueling process. If the deviation between the two exceeds the preset allowable range, the annotation information is corrected based on the real-time recorded data.

[0067] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0068] This invention provides a method for constructing a sample database for aerial refueling scenarios. During execution, this method ensures comprehensive data coverage and relevance to actual aerial refueling scenarios by subdividing meteorological environment categories and accurately calculating influence coefficients. Data collection involves multi-dimensional screening to remove inferior data, ensuring data quality. Image frames are extracted at different time intervals according to refueling stages and synchronized with precise timestamps, making the data more timely and targeted. Data format conversion follows unified standards and verifies integrity, improving data consistency and usability. Adding annotation information combines real-time perception and preprocessing from multiple sensors to accurately correlate meteorological parameters. Through feature extraction, localization, and stage determination models, component locations and refueling stages are accurately labeled. A hierarchical storage architecture and multi-level indexes are constructed to optimize data management and access efficiency. Data is verified, supplemented, and corrected to further ensure database integrity and accuracy, providing effective data support for aerial refueling-related research and algorithm training. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0070] Figure 1 This is a flowchart illustrating a method for constructing a sample database for an aerial refueling scenario. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0072] The present invention will be further described below with reference to embodiments.

[0073] Example:

[0074] This embodiment presents a method for constructing a sample database for an aerial refueling scenario, such as... Figure 1 As shown, it includes:

[0075] Set up extreme, normal, and special weather environment categories, and collect corresponding video and image data of the aerial refueling process based on each category;

[0076] Extreme, normal, and special meteorological environment categories shall be set according to the following:

[0077] Obtain meteorological parameters, calculate the dynamic coupling influence coefficient of meteorological parameters on aerial refueling safety, and determine the meteorological environment category based on the influence coefficient:

[0078] ;

[0079] Among them, the single-factor nonlinear response function is:

[0080] ;

[0081] Two-factor coupling function:

[0082] ;

[0083] Time-domain rate of change function:

[0084] ;

[0085] Relative position safety margin correction function:

[0086] ;

[0087] In the formula: This represents the dynamic coupling influence coefficient. These correspond to four meteorological factors: wind speed, precipitation, temperature difference, and visibility. The basic weight coefficient for the p-th meteorological factor; For the p-th and q-th meteorological factors, the coupling weight coefficient is denoted as . This is the weighting coefficient for the rate of change in the time domain; This refers to the relative position safety margin weighting coefficient; Let p be the measured value of the p-th meteorological factor; The safety threshold for the p-th meteorological factor; This is the reference standard value for the p-th meteorological factor; , Let p be the nonlinear response parameter of the p-th meteorological factor. Used to control the intensity of nonlinear growth, Control the amplification rate after exceeding the threshold; , Let p and q be the coupling strength coefficients of the meteorological factors. Reflecting the degree of interaction between different meteorological factors, Controlling the degree of nonlinearity of coupling; , , Let be the time-domain dynamic parameter of the p-th factor. Reflecting the relative importance of the rate of change of each meteorological factor, Controlling the sensitivity of the response, Used to adjust the degree of nonlinearity of the effect of the rate of change; These are the relative position parameters of the two machines; The optimal relative position parameters; For the current safety margin; The critical safety margin is determined based on the dual-unit models and refueling method; The standard deviation of the relative position tolerance; Adjust the weighting coefficients for the location;

[0088] It should be noted that traditional meteorological impact assessments for aerial refueling use simple linear weighted models, which cannot accurately reflect the nonlinear coupling effects, threshold effects, and temporal dynamic characteristics among multiple meteorological factors. This results in insufficient prediction accuracy under complex meteorological conditions, especially when multiple meteorological parameters simultaneously approach or exceed safety thresholds. In such cases, the linear model severely underestimates the comprehensive impact of meteorological conditions on aerial refueling safety. The dynamic coupling impact coefficient model described in this invention decomposes meteorological impacts into four mutually coupled components: a single-factor nonlinear response uses a piecewise function to introduce an exponential amplification effect after exceeding the safety threshold, based on the nonlinear response characteristics of meteorological parameters in atmospheric turbulence theory; a two-factor coupling function uses a combination of power and sine functions based on the multi-parameter interaction mechanism in atmospheric dynamics to reflect the resonance amplification effect among meteorological factors; a temporal rate of change function uses a hyperbolic tangent function to describe the saturation impact characteristics of the meteorological parameter change rate; and a relative position safety margin correction function uses a Gaussian function and an exponential decay term to reflect the moderating effect of the geometric position deviation between the two aircraft on meteorological sensitivity.

[0089] The parameters and coefficients in the model are determined based on actual aerial refueling scenarios, aircraft combinations, and safety requirements. In the K formula, the sum of the weights of each calculation term should be 1 to ensure model normalization. The weight coefficients γ for the time-domain rate of change and δ for the relative position safety margin should be less than the basic weights to maintain the dominant factor status. , To ensure over-threshold amplification effect, time-domain parameters It should be matched with the physical response time of each meteorological factor;

[0090] when When the temperature is ≥8, the meteorological environment is classified as an extreme meteorological environment; when When ≤3, the meteorological environment is classified as a routine meteorological environment; when 3 < If the weather is below 8:00, the meteorological environment will be classified as a special meteorological environment.

[0091] The above formula transforms multi-dimensional meteorological parameters into a single coefficient that can be directly used to determine the category, which solves the problem of isolated parameters and difficulty in comprehensive evaluation in traditional meteorological classification. Furthermore, by customizing the frequency standard of sudden meteorological changes under special meteorological conditions, it further refines the classification of meteorological categories, providing an accurate basis for subsequent data collection and storage by meteorological category, and ensuring that the database data covers meteorological scenarios with different safety risk levels.

[0092] Among them, the special meteorological environment category also includes sudden meteorological changes with a frequency of ≥2 times / hour. Sudden meteorological changes are defined as a single meteorological parameter change exceeding 50% of the parameter's normal fluctuation range. The normal fluctuation range of the parameter is defined by the user.

[0093] Filter the collected video and image data, and remove data that is blurry, lacks key targets, or has excessive shooting errors;

[0094] The filtering and removal processes for the collected video and image data include:

[0095] Calculate the resolution evaluation index of video or image. , Indicates the height and width in pixels of a video frame or image. This represents the gradient value of pixel (i,j) in the x-direction. This represents the gradient value of pixel (i,j) in the y-direction. < At that time, the image was determined to be blurry and was removed. A preset resolution threshold is set, with a value between 50 and 80, which can be customized by the user. Higher resolution video and image data will be subject to stricter quality requirements. The larger the value;

[0096] The above formula constructs an objective evaluation index by quantifying gradient information, and The threshold can be flexibly adjusted according to actual data quality requirements, taking into account data screening standards in different scenarios, effectively removing blurry data, ensuring the clarity and consistency of image data in the database, and laying a high-quality data foundation for subsequent key target identification, component positioning and other operations.

[0097] A pre-defined set of key targets for aerial refueling is used, including the refueling probe of the tanker aircraft, the refueling port of the receiver aircraft, the fuselage markings of the tanker aircraft and the receiver aircraft. Target detection algorithms are used to identify key targets in videos or images. If any key target is missing in the identification results, it is determined to be a missing key target and is removed.

[0098] Calculate the shooting angle deviation value And the shooting distance deviation value D, when > And D > If the shooting error exceeds the standard, it will be rejected.

[0099] in, The preset viewing angle deviation threshold is set to 5-8 degrees. The preset distance deviation threshold is set to 5m-10m. The angle between the actual shooting angle and the standard shooting angle is represented by , and D represents the difference between the actual shooting distance and the standard shooting distance.

[0100] Extract image frames from the filtered video at fixed time intervals, and convert the image frames and the original image data into a preset standard format;

[0101] The operation of extracting image frames from the filtered video at fixed time intervals includes:

[0102] The refueling process is divided into four phases: preparation and docking, precise docking, stable refueling, and separation and withdrawal. Fixed time intervals are set for each phase.

[0103] The fixed time interval for the preparation docking phase is set to 2s to 3s; the fixed time interval for the precise docking phase is set to 0.5s to 1s; the fixed time interval for the stable refueling phase is set to 3s to 5s; and the fixed time interval for the separation and withdrawal phase is set to 2s to 3s.

[0104] When extracting image frames, the timestamp corresponding to each image frame is recorded synchronously, and the timestamp is accurate to the millisecond level and kept in time synchronization with the real-time clock system of the aerial refueling scenario.

[0105] It should be noted that:

[0106] The preparation docking phase is the process by which the refueling aircraft and the receiving aircraft approach each other from a distance to the preset docking preparation position. The relative motion speed of the two sides is stable and the distance changes gradually, so the fixed time interval is set to 2s to 3s.

[0107] The precise docking phase is the process from the receiving aircraft adjusting from the docking preparation position to the coupling of the refueling probe and the receiving port. The relative attitude of the two sides is adjusted frequently and the distance is reduced rapidly. Therefore, the fixed time interval is set to 0.5s to 1s.

[0108] The stable refueling phase is the process in which the refueling probe and the fuel receiving port maintain a coupled state and transfer fuel. The relative position and attitude of the two remain stable, so the fixed time interval is set to 3s to 5s.

[0109] The separation and withdrawal phase is the process in which the refueling probe separates from the refueling port after refueling is completed and gradually increases the distance between them. The relative speed of the two sides changes from static to dynamic and the distance increases slowly. Therefore, the fixed time interval is set to 2s to 3s.

[0110] Add meteorological environmental parameters, locations of key components for aerial refueling, and refueling phase annotations to the converted data.

[0111] When converting image frames and raw image data to a preset standard format, the following applies:

[0112] The preset standard format parameters include an image resolution of 1920×1080 pixels, a color space of RGB, an image compression format of JPEG, and a compression quality factor of 85-90;

[0113] During the conversion process, an interpolation algorithm is used to adjust the resolution of non-standard resolution image frames and original image data. A color space conversion matrix suitable for aerial refueling scenarios is used to convert the non-RGB color space image data to color space. The color space conversion matrix meets the following conditions:

[0114] ;

[0115] In the formula: R, G, and B are the red, green, and blue component values ​​of the RGB color space, respectively; Y, U, and V are the luminance, blue difference, and red difference component values ​​of the YUV color space, respectively.

[0116] After the conversion is completed, the integrity of the image data is checked. If the check finds that the image data is damaged or missing, the conversion operation is re-executed.

[0117] It should be noted that:

[0118] The parameters in the decimal part of the above formula are determined based on the precise color reproduction requirements of key targets (such as the refueling aircraft probe and the receiving aircraft's refueling port) in the image data of the aerial refueling scenario, combined with the human visual system's perception sensitivity to the three primary colors of red, green and blue, and the physical mapping relationship between brightness (Y), blue difference (U), red difference (V) and RGB components in the YUV color space.

[0119] The contribution coefficients of brightness Y to RGB (0.299, 0.587, 0.114) are based on the internationally recognized Rec.601 standard, which matches the visual characteristics of the human eye, which is most sensitive to green, followed by red, and least sensitive to blue, ensuring that the brightness information of the converted image is consistent with the original scene. The coefficients of U and V in the green (G) and blue (B) components are obtained through iterative optimization by testing a large number of image samples under different lighting conditions (such as strong light at high altitudes and weak light due to cloud cover) in aerial refueling scenarios, with the goal of "color deviation of key components <3%". This ensures the color fidelity when converting non-RGB formats (such as YUV) to RGB formats, and avoids errors in subsequent key component location marking and refueling stage judgment due to color distortion, fully adapting to the special requirements of aerial refueling scenarios for the color accuracy of image data.

[0120] This matrix is ​​constructed based on the color mapping relationship between RGB and YUV color spaces. It uses a linear transformation to convert the red (R), green (G), and blue (B) components of the RGB color space into the luminance (Y), blue difference (U), and red difference (Y) components of the YUV color space. The luminance (Y) is calculated from the three RGB components with specific weights, while the color differences (U and V) are calculated from the differences between the RGB components and Y. This conforms to the general color science principles of color space conversion, ensuring that the color information of the converted image is not distorted. It solves the problem of incompatibility between data from different color spaces and performs a simultaneous integrity check after conversion to prevent data corruption, ensuring that all image data in the database has a consistent color format and providing a unified color benchmark for subsequent feature extraction, annotation, and other operations.

[0121] When adding meteorological environmental parameter annotation information to the converted data, real-time meteorological environmental parameters are obtained by sensing through the air-ground cooperative sensor network installed on the refueling aircraft and the receiving aircraft. The air-ground cooperative sensor network includes wind speed sensors, precipitation sensors, temperature sensors, and visibility sensors, with a sampling frequency of no less than 1 time / second.

[0122] Simultaneously, the collected real-time meteorological and environmental parameters are preprocessed, including outlier removal and adaptive smoothing. The outlier removal operation is as follows:

[0123] When any collected data satisfy When an outlier is detected, it is identified and removed. This is the mean of the data collected for this parameter. The standard deviation of the data collected for this parameter;

[0124] The adaptive smoothing process is as follows:

[0125] ;

[0126] In the formula: The value of the meteorological parameter at time t is the smoothed value. To adapt to window size; As a time decay weight; represents the raw values ​​of meteorological parameters collected at time k; a and b are adjustment coefficients;

[0127] The above formula associates the window size with the frequency of meteorological parameter fluctuations, avoiding the problems of the fixed window moving average method, which may result in excessive smoothing and loss of key change information when parameter fluctuations are drastic, or insufficient smoothing and retention of too much noise when fluctuations are gentle. At the same time, combined with the previous outlier removal operation, it ensures that the preprocessed meteorological parameters are accurate and stable, providing a reliable meteorological annotation basis for subsequent time correlation matching with image data.

[0128] Then, the preprocessed meteorological and environmental parameters are matched with the converted data in time, so that each image frame and the original image data correspond to a unique set of meteorological and environmental parameter annotation information.

[0129] Among them, adaptive window size ∈[5,10], when the real-time fluctuation frequency of meteorological parameters in the aerial refueling scenario is low and the data changes smoothly, The larger the value, the more frequent the real-time fluctuations of meteorological parameters and the more drastic the data changes. The smaller the value;

[0130] When adding location annotations for key aerial refueling components to the converted data, a feature extraction and localization model for these components is simultaneously constructed. This model includes a component feature enhancement module, a multi-scale feature matching module, and a location coordinate optimization module.

[0131] Component feature enhancement module: Performs grayscale layering processing on the format-converted image frames and the original image data, dividing the image grayscale values ​​into 16 continuous grayscale intervals; calculates the texture density and edge gradient ratio of key aerial refueling components for each grayscale interval;

[0132] Gray-scale intervals with a texture density greater than or equal to 0.6 and an edge gradient ratio greater than or equal to 0.5 are selected as key feature intervals. The gray-scale values ​​of pixels in the key feature intervals are linearly stretched so that the stretched gray-scale value range is mapped to 128-255. The gray-scale values ​​of pixels in non-key feature intervals are linearly compressed so that the compressed gray-scale value range is mapped to 0-64.

[0133] The texture density is determined by the ratio of the number of texture pixels of key components in the interval to the total number of pixels in the interval, and the edge gradient ratio is determined by the ratio of the number of pixels whose edge pixel gradient values ​​of key components in the interval are greater than or equal to a preset gradient threshold to the total number of pixels of key components in the interval.

[0134] Multi-scale feature matching module: A standard feature template library for the refueling probe of the tanker aircraft and the refueling port of the receiver aircraft is pre-set. The template library contains key component templates corresponding to no fewer than three pre-set combinations of aerial refueling aircraft models. Each template includes standard feature samples under at least five different attitude angles and three different lighting intensities within the range of 0-30 degrees. The image features processed for key feature intervals are matched against the standard feature template library at multiple scales to calculate feature similarity. The formula is as follows:

[0135] S = ∑ m = 1 M ∑ n = 1 N [ F ( m , n ) × T ( m , n ) ] ∑ m = 1 M ∑ n = 1 N F ( m , n ) 2 × ∑ m = 1 M ∑ n = 1 N T ( m , n ) 2 ;

[0136] In the formula: These are the height and width pixels of the feature template, respectively. It is the feature value at position (m,n) within the key feature interval of the image, that is, the gray value at that position after gray-scale stretching; For the feature value at position (m,n) in the standard feature template, i.e., the standard gray value at the corresponding position in the template, select the similarity. The image area corresponding to the standard template with a value ≥0.85 is used as the initial positioning area for key components;

[0137] The above formula calculates the sum of the absolute differences between the feature values ​​in the key feature interval of the image and the corresponding feature values ​​in the standard feature template, and then divides it by the product of the template height and width for normalization to obtain the feature similarity S. The closer S is to 1, the more similar the image features are to the standard template. When S≥0.85, it is determined as the preliminary location area of ​​the key component. The principle is that the absolute difference can intuitively reflect the degree of difference between the two features. The smaller the difference, the higher the similarity.

[0138] Furthermore, the above formula is based on the requirement of multi-scale feature matching, constructs a quantitative similarity evaluation standard, and the standard feature template library covers a variety of aircraft combinations, attitude angles and illumination intensities, which solves the problem of single templates and difficulty in adapting to complex aerial refueling scenarios in traditional feature matching. By setting a similarity threshold of 0.85, the preliminary positioning area that meets the requirements is accurately selected, providing an accurate initial range for subsequent optimization of the position coordinates of key components and improving the positioning accuracy of components.

[0139] Position coordinate optimization module: Based on the initial positioning area, the region boundary is scanned using the 8-neighbor pixel traversal method, with the traversal step size set to 1 pixel, to determine the minimum bounding rectangle of the key components.

[0140] Using the top-left corner of the image as the origin, with the positive x-axis pointing horizontally to the right and the positive y-axis pointing vertically downwards, record the coordinates of the top-left corner of the smallest bounding rectangle. and the coordinates of the bottom right corner ;

[0141] Calculate the relative positional parameters of key components, including the straight-line distance between the refueling probe of the fuel dispenser and the refueling port of the receiving unit. and angle deviation The calculation formula is:

[0142] ;

[0143] In the formula: ( ( ) represents the center coordinates of the rectangular bounding box of the fuel dispenser's fuel probe; () represents the center coordinates of the rectangular bounding box of the receiver's oil inlet; Standard angles for aligning key components during aerial refueling;

[0144] Add the location coordinates and relative positional parameters of key components as location annotation information to the corresponding data;

[0145] Among them, key components include at least the refueling probe of the refueling machine and the oil receiving port of the receiving machine. The three light intensity settings are low light 50-100 lux, normal light 100-500 lux, and strong light 500-1000 lux. The matching scale in the multi-scale matching operation is set to 0.8 times, 1.0 times, and 1.2 times.

[0146] The above formula expands the location of key components from single coordinate labeling to the quantification of relative positional relationships, which solves the problem that traditional location labeling can only reflect the location of a single component and cannot reflect the coordination status between components. Furthermore, by using two parameters, distance and angular deviation, it accurately describes the alignment accuracy of components, providing information on the coordination status of key components for refueling stage judgment and database data labeling, making the labeled data more in line with the actual operational needs of aerial refueling.

[0147] When adding refueling stage annotation information to the converted data, the refueling stage determination model is constructed through the following steps:

[0148] Feature Parameter Selection and Preprocessing: Core feature parameters during the aerial refueling process are selected as model input variables. These core feature parameters include the real-time relative distance d between the tanker and the receiver aircraft, the real-time relative speed v, and the alignment accuracy of key components. ;

[0149] The selected feature parameters are standardized using the following formula: , These are the standardized feature parameter values. These are the original values ​​of the feature parameters. These are the maximum and minimum values ​​of the feature parameter in the historical aerial refueling dataset, respectively.

[0150] The determination logic is set as follows:

[0151] When the standardized relative distance exceeds 0.8 and the standardized relative speed exceeds 0.7, it is determined to be in the docking preparation stage. Here, a standardized relative distance exceeding 0.8 corresponds to an actual relative distance greater than 500m, and a standardized relative speed exceeding 0.7 corresponds to an actual relative speed greater than 10m / s.

[0152] When the standardized relative distance is in the range of [0.1, 0.8], the standardized relative speed is in the range of [0.2, 0.7], and the alignment accuracy of key components is less than 90%, it is judged to be in the precise docking stage. Among them, the standardized relative distance in the range of [0.1, 0.8] corresponds to the actual relative distance in the range of [10m, 500m], and the standardized relative speed in the range of [0.2, 0.7] corresponds to the actual relative speed in the range of [0.5m / s, 10m / s].

[0153] When the standardized relative distance is less than 0.1, the standardized relative speed is less than 0.2, and the alignment accuracy of key components is not less than 90%, it is determined to be a stable refueling stage. Here, a standardized relative distance of less than 0.1 corresponds to an actual relative distance of less than 10m, and a standardized relative speed of less than 0.2 corresponds to an actual relative speed of less than 0.5m / s.

[0154] When the standardized relative distance exceeds 0.3, the standardized relative speed exceeds 0.5, and the alignment accuracy of key components is <50%, it is determined to be the separation and withdrawal stage. Among them, the standardized relative distance exceeding 0.3 corresponds to the actual relative distance exceeding 100m, and the standardized relative speed exceeding 0.5 corresponds to the actual relative speed exceeding 5m / s.

[0155] Model validation and optimization: The constructed judgment model is validated using a pre-set historical aerial refueling annotation dataset. The stage judgment accuracy of the model is calculated. If the accuracy is lower than 98%, the judgment threshold range of each stage is adjusted based on the feature parameter distribution of the erroneous judgment samples until the model accuracy is not less than 98%.

[0156] The output of the optimized refueling stage determination model is used as the refueling stage labeling information and added to the corresponding data. If there is a fuzzy area where the threshold intervals of two stages overlap in the model determination result, the real-time aerial refueling operation record corresponding to the data is manually retrieved for verification to determine the final refueling stage labeling.

[0157] in, , This indicates the angular deviation between the fuel dispenser's fuel probe and the receiving port of the fuel dispenser;

[0158] Build a database infrastructure, classify and store labeled data according to meteorological environment category and refueling stage, and simultaneously establish related indexes;

[0159] During the database infrastructure construction phase, labeled data is categorized and stored according to meteorological environment type and refueling stage, and related indexes are established simultaneously:

[0160] Database infrastructure includes a data storage layer, an index management layer, and an access interface layer;

[0161] Data storage layer:

[0162] The system adopts a segmented storage approach, dividing the data into extreme weather storage segments, regular weather storage segments, and special weather storage segments according to meteorological environment categories. Within each meteorological storage segment, sub-storage areas are synchronously divided based on the refueling stage. Labeled data is stored in the corresponding sub-storage areas according to the corresponding meteorological environment category and refueling stage. During storage, the data is encrypted using a custom encryption algorithm.

[0163] Index Management Layer:

[0164] Set up a multi-level associated index: the first-level index uses the meteorological environment category as the index key, the second-level index uses the refueling stage as the index key, and the third-level index uses the data collection timestamp as the index key. The index structure adopts a B+ tree index structure, and the order of the B+ tree index is set to 100-200.

[0165] Access interface layer:

[0166] The data access interface is configured to support both SQL queries and API calls, with the interface response time controlled to be less than 100ms. Different data access permissions are also set for different user roles.

[0167] The data volume and completeness of the annotation entries for each meteorological environment category in the database are verified one by one. The aerial refueling video and image data of the category with insufficient data volume are supplemented, and the meteorological parameters, component locations and stage division contents in the annotation information that do not match the actual scene are corrected.

[0168] The database validation, completion, and correction operations are as follows:

[0169] When verifying the amount of data, preset minimum thresholds for video and image data under each meteorological environment category are used. The minimum threshold for video data under extreme meteorological environment category is 500 hours and the minimum threshold for image data is 100,000 frames. The minimum threshold for video data under normal meteorological environment category is 2,000 hours and the minimum threshold for image data is 500,000 frames. The minimum threshold for video data under special meteorological environment category is 1,000 hours and the minimum threshold for image data is 200,000 frames.

[0170] If the amount of data under a certain meteorological environment category is lower than the corresponding minimum threshold, the data completion process is initiated: targeted data collection is carried out by adjusting the acquisition parameters and acquisition duration of the air-ground collaborative sensor network until the amount of data meets the minimum threshold requirement.

[0171] When verifying the completeness of labeled entries, calculate the label completeness evaluation index:

[0172] , To label the integrity evaluation indicators, To determine the number of entries with accurate and complete information. The total number of all labeled entries in the database, when If the percentage is less than 95%, the item is deemed to be incomplete and the incomplete item needs to be manually completed.

[0173] When correcting the labeling information, the meteorological parameters, component locations, and phase divisions in the labeling information are compared with the real-time recorded data of the aerial refueling process. If the deviation between the two exceeds the preset allowable range, the labeling information is corrected based on the real-time recorded data.

[0174] The preset allowable ranges include: meteorological parameter deviation not exceeding 10%, component position coordinate deviation not exceeding 5 pixels, and stage division time deviation not exceeding 1 second.

[0175] In this embodiment, the above method can accurately collect aerial refueling data under various weather conditions. After screening, standardization, and standardized labeling, a high-quality classification database is constructed. Its multi-level index and permission management ensure efficient and secure data access, and the data verification and completion mechanism ensures sufficient and complete data. Moreover, it can meet the needs of aerial refueling-related algorithm training, scenario simulation, etc., providing reliable data support for improving refueling safety and optimizing operation processes, and promoting the efficiency improvement of aerial refueling technology research and application.

[0176] In summary, the methods described in the above embodiments, during execution, ensure comprehensive data coverage and relevance to actual aerial refueling scenarios by subdividing meteorological environment categories and accurately calculating influence coefficients. Data quality is guaranteed by multi-dimensional screening to remove inferior data during collection. Image frames are extracted at different time intervals according to the refueling stage and synchronized with precise timestamps, making the data more timely and targeted. Data format conversion follows unified standards and verifies integrity, improving data consistency and usability. Adding annotation information combines real-time perception and preprocessing from multiple sensors to accurately correlate meteorological parameters. Through feature extraction, localization, and stage determination models, component locations and refueling stages are accurately labeled. A hierarchical storage architecture and multi-level indexes are constructed to optimize data management and access efficiency. Data is verified, supplemented, and corrected to further ensure database integrity and accuracy, providing effective data support for aerial refueling-related research and algorithm training.

[0177] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a sample database for aerial refueling scenarios, characterized in that, include: Meteorological parameters are acquired, and the dynamic coupling influence coefficient of meteorological parameters on aerial refueling safety is calculated. The meteorological environment category is determined by referring to the dynamic coupling influence coefficient. When the dynamic coupling influence coefficient is ≥8, it is set as an extreme meteorological environment category. When the dynamic coupling influence coefficient is ≤3, it is set as a normal meteorological environment category. When 3 < dynamic coupling influence coefficient < 8, it is set as a special meteorological environment category. Based on each category, corresponding video and image data of the aerial refueling process are collected. Filter the collected video and image data, and remove data that is blurry, lacks key targets, or has excessive shooting errors; Extract image frames from the filtered video at fixed time intervals, and convert the image frames and the original image data into a preset standard format; Meteorological parameters, locations of key aerial refueling components, and refueling phase annotations are added to the format-converted data. While adding these annotations, a feature extraction and localization model for key aerial refueling components is simultaneously constructed. This model includes a component feature enhancement module, a multi-scale feature matching module, and a position coordinate optimization module. The component feature enhancement module performs grayscale layering on the format-converted image frames and the original image data, filters key feature intervals, and enhances key component features through linear stretching and compression of grayscale values. The multi-scale feature matching module pre-sets a standard feature template library for the refueling probe of the tanker aircraft and the refueling port of the receiver aircraft, calculates the similarity between the key component features and each standard feature template in the library, and selects the initial location area for the key component based on the similarity. The position coordinate optimization module uses the initial location area as a reference, scans the area boundary using an 8-neighbor pixel traversal method, determines the minimum bounding rectangle of the key component, and calculates the relative positional relationship parameters of the key component. Build a database infrastructure, classify and store labeled data according to meteorological environment category and refueling stage, and simultaneously establish related indexes; The data volume and completeness of the annotation entries for each meteorological environment category in the database are verified one by one. The aerial refueling video and image data of the category with insufficient data volume are supplemented, and the meteorological parameters, component locations and stage division contents in the annotation information that do not match the actual scene are corrected. The key targets include the refueling probe of the refueling machine, the oil receiving port of the receiving machine, the markings on the body of the refueling machine, and the markings on the body of the receiving machine. The key components include at least the refueling probe of the refueling machine and the oil receiving port of the receiving machine.

2. The method for constructing a sample database for aerial refueling scenarios according to claim 1, characterized in that, The dynamic coupling influence coefficient is calculated using the following formula: ； Among them, the single-factor nonlinear response function is: ； Two-factor coupling function: ； Time-domain rate of change function: ； Relative position safety margin correction function: ； In the formula: This represents the dynamic coupling influence coefficient. These correspond to four meteorological factors: wind speed, precipitation, temperature difference, and visibility. The basic weight coefficient for the p-th meteorological factor; For the p-th and q-th meteorological factors, the coupling weight coefficient is denoted as . This is the weighting coefficient for the rate of change in the time domain; This refers to the relative position safety margin weighting coefficient; Let p be the measured value of the p-th meteorological factor; The safety threshold for the p-th meteorological factor; This is the reference standard value for the p-th meteorological factor; , Let be the nonlinear response parameter of the p-th meteorological factor; , Let be the coupling strength coefficient between the p-th and q-th meteorological factors; , , Let be the time-domain dynamic parameter of the p-th factor; These are the relative position parameters of the two machines; The optimal relative position parameters; For the current safety margin; This represents the critical safety margin. The standard deviation of the relative position tolerance; Adjust the weighting coefficients for the location.

3. The method for constructing a sample database for aerial refueling scenarios according to claim 1, characterized in that, The video and image data collected through screening are filtered to remove data that is blurry, lacks key targets, or has excessive shooting errors, including: Calculate the resolution evaluation index of video or image. , Indicates the height and width in pixels of a video frame or image. This represents the gradient value of pixel (i,j) in the x-direction. This represents the gradient value of pixel (i,j) in the y-direction. < At that time, the image was determined to be blurry and was removed. Set a preset resolution threshold; A pre-defined set of key targets for aerial refueling is used, including the refueling probe of the tanker aircraft, the refueling port of the receiver aircraft, the fuselage markings of the tanker aircraft and the receiver aircraft. Target detection algorithms are used to identify key targets in videos or images. If any key target is missing in the identification results, it is determined to be a missing key target and is removed. Calculate the shooting angle deviation value And the shooting distance deviation value D, when > And D > If the shooting error exceeds the standard, it will be rejected. in, To preset the viewing angle deviation threshold, To preset the distance deviation threshold, The angle between the actual shooting angle and the standard shooting angle is represented by , and D represents the difference between the actual shooting distance and the standard shooting distance.

4. The method for constructing a sample database for an aerial refueling scenario according to claim 1, characterized in that, The operation of extracting image frames from the filtered video at fixed time intervals includes: The refueling process is divided into four phases: preparation and docking, precise docking, stable refueling, and separation and withdrawal. Fixed time intervals are set for each phase. When extracting image frames, the timestamp corresponding to each image frame is recorded synchronously. The timestamp is accurate to the millisecond level and is synchronized with the real-time clock system of the aerial refueling scenario.

5. The method for constructing a sample database for an aerial refueling scenario according to claim 1, characterized in that, When converting image frames and raw image data into a preset standard format, the preset standard format parameters include image resolution, color space, and image compression format. During the conversion process, interpolation algorithms are used to adjust the resolution of non-standard resolution image frames and original image data, and a color space conversion matrix adapted to the aerial refueling scenario is used to convert the non-RGB color space image data to color space. ； In the formula: R, G, and B are the red, green, and blue component values ​​of the RGB color space, respectively; Y, U, and V are the luminance, blue difference, and red difference component values ​​of the YUV color space, respectively. After the conversion is completed, the image data is checked for integrity. If the check finds that the image data is damaged or missing, the conversion operation is repeated.

6. The method for constructing a sample database for aerial refueling scenarios according to claim 1, characterized in that, When adding meteorological parameter annotation information to the converted data, real-time meteorological parameters are obtained by sensing through an air-ground cooperative sensor network installed on the refueling aircraft and the receiving aircraft. The air-ground cooperative sensor network includes wind speed sensors, precipitation sensors, temperature sensors, and visibility sensors. The collected real-time meteorological parameters are preprocessed synchronously, including outlier removal and adaptive smoothing. The outlier removal operation is as follows: When any collected data satisfy When an outlier is detected, it is identified and removed. This is the average value of the collected data for this meteorological parameter. The standard deviation of the collected data for this meteorological parameter; The adaptive smoothing process is as follows: ； In the formula: The value of the meteorological parameter at time t is the smoothed value. To adapt to window size; As a time decay weight; These are the original values ​​of the meteorological parameters collected at time k; a and b are adjustment coefficients; The preprocessed meteorological parameters are matched with the format-converted data over time, so that each image frame and the original image data correspond to a unique set of meteorological parameter annotation information.

7. The method for constructing a sample database for aerial refueling scenarios according to claim 1, characterized in that, The component feature enhancement module performs grayscale layering processing on the format-converted image frames and the original image data, and selects grayscale intervals with texture density greater than or equal to 0.6 and edge gradient ratio greater than or equal to 0.5 as key feature intervals. The key component features are enhanced by linear stretching and compression of grayscale values. The multi-scale feature matching module uses a pre-defined feature template library of refueling probes and receiving ports of fuel dispensers to calculate feature similarity. ； In the formula: These are the height and width pixels of the feature template, respectively. It is the feature value at position (m,n) within the key feature interval of the image, that is, the gray value at that position after gray-scale stretching; For the feature value at position (m,n) in the standard feature template, i.e., the standard gray value at the corresponding position in the template, select the similarity. The image area corresponding to the standard template with a value ≥0.85 is used as the initial positioning area for key components; Calculate the relative positional parameters of key components, including the straight-line distance between the refueling probe of the fuel dispenser and the refueling port of the receiving unit. and angle deviation The calculation formula is: ； In the formula: ( ( ) represents the center coordinates of the rectangular bounding box of the fuel dispenser's fuel probe; () represents the center coordinates of the rectangular bounding box of the receiver's oil inlet; Standard angles for aligning key components during aerial refueling; The location coordinates and relative positional parameters of key components are added to the corresponding data as location annotation information.

8. A method for constructing a sample database for an aerial refueling scenario according to claim 4, characterized in that, When adding refueling stage annotation information to the converted data, the refueling stage determination model is constructed through the following steps: The core feature parameters of the aerial refueling process are selected as the model input variables. These core feature parameters include the real-time relative distance d between the tanker and the receiver aircraft, the real-time relative speed v, and the alignment accuracy of key components. ; Standardize the selected feature parameters; The determination logic is set as follows: When the standardized relative distance exceeds 0.8 and the standardized relative speed exceeds 0.7, it is determined to be in the docking preparation stage. Here, a standardized relative distance exceeding 0.8 corresponds to an actual relative distance greater than 500m, and a standardized relative speed exceeding 0.7 corresponds to an actual relative speed greater than 10m / s. When the standardized relative distance is in the range of [0.1, 0.8], the standardized relative speed is in the range of [0.2, 0.7], and the alignment accuracy of key components is less than 90%, it is judged to be in the precise docking stage. Among them, the standardized relative distance in the range of [0.1, 0.8] corresponds to the actual relative distance in the range of [10m, 500m], and the standardized relative speed in the range of [0.2, 0.7] corresponds to the actual relative speed in the range of [0.5m / s, 10m / s]. When the standardized relative distance is less than 0.1, the standardized relative speed is less than 0.2, and the alignment accuracy of key components is not less than 90%, it is determined to be a stable refueling stage. Here, a standardized relative distance of less than 0.1 corresponds to an actual relative distance of less than 10m, and a standardized relative speed of less than 0.2 corresponds to an actual relative speed of less than 0.5m / s. When the standardized relative distance exceeds 0.3, the standardized relative speed exceeds 0.5, and the alignment accuracy of key components is <50%, it is determined to be the separation and withdrawal stage. Among them, the standardized relative distance exceeding 0.3 corresponds to the actual relative distance exceeding 100m, and the standardized relative speed exceeding 0.5 corresponds to the actual relative speed exceeding 5m / s. The constructed judgment model is validated using a pre-set historical aerial refueling annotation dataset. If the accuracy is lower than 98%, the judgment threshold range for each stage is adjusted.

9. A method for constructing a sample database for an aerial refueling scenario according to claim 1, characterized in that, The database infrastructure includes a data storage layer, an index management layer, and an access interface layer: The data storage layer adopts a segmented storage method, which is divided into extreme weather storage segments, normal weather storage segments, and special weather storage segments according to meteorological environment categories. Each meteorological storage segment is further divided into sub-storage areas based on the refueling stage. The index management layer is set up with multi-level related indexes: the first-level index uses the meteorological environment category as the index key, the second-level index uses the refueling stage as the index key, and the third-level index uses the data collection timestamp as the index key. The index structure adopts a B+ tree index structure. The access interface layer is configured to support both SQL queries and API calls for data access.

10. A method for constructing a sample database for an aerial refueling scenario according to claim 1, characterized in that, The verification, completion, and correction operations in the database are as follows: When verifying the data volume, a minimum threshold for the video and image data volume under each meteorological environment category is preset. If the data volume under a certain meteorological environment category is lower than the corresponding minimum threshold, the data completion process is initiated. When verifying the completeness of labeled entries, calculate the label completeness evaluation index: , To label the integrity evaluation indicators, To determine the number of entries with accurate and complete information. The total number of all labeled entries in the database, when If the percentage is less than 95%, the item is deemed to be incomplete and the incomplete item needs to be manually completed. When correcting the annotation information, the meteorological parameters, component locations, and phase divisions in the annotation information are compared with the real-time recorded data of the aerial refueling process. If the deviation between the two exceeds the preset allowable range, the annotation information is corrected based on the real-time recorded data.

Images