Photovoltaic construction progress intelligent control method and system based on flight image analysis

Abstract

The invention provides a photovoltaic construction progress intelligent management and control method and system based on flight image analysis. The system comprises a central server, an unmanned aerial vehicle airport system and an edge computing terminal, the method sequentially comprises the following steps: importing design data and a construction plan, planning a flight area and a path, issuing a server task and executing an airport, automatically flight by an unmanned aerial vehicle and acquiring an image, and establishing a spatial association relationship between an image and a corresponding construction grid by adopting a non-orthographic grid matching method. And the edge computing terminal performs image recognition processing and grid progress statistics and visual display. According to the method, photovoltaic design data, a construction plan and unmanned aerial vehicle aerial images are combined, automatic identification, space positioning and gridding progress statistics of the photovoltaic module and the support are achieved, accumulation and increment calculation can be conducted on the construction completion condition of each grid unit, the authenticity and timeliness of progress data are ensured, and meanwhile, the construction efficiency is improved. The platform can generate a visual progress report, and low-intervention and intelligent construction management is realized.

Description

TECHNICAL FIELD

[0001] The application relates to an intelligent photovoltaic construction progress management method and system based on aerial image analysis, and belongs to the technical research field of construction engineering software. BACKGROUND

[0002] In recent years, photovoltaic power station construction has rapidly developed worldwide, especially in centralized and large-scale ground power station projects. The installation scale of photovoltaic modules is huge, and the construction period is tight, which puts forward higher requirements for the fine management of project progress. During the construction process, the daily module installation quantity, installation position and completion status and other information are directly related to the overall production scheduling, dispatching and acceptance node of the project. Therefore, how to efficiently and accurately obtain and master the actual installation progress of the photovoltaic module has become a key link in the digital construction management of the photovoltaic project.

[0003] At present, most photovoltaic projects still rely on manual inspection and manual reporting to obtain progress information. Construction management personnel usually perform on-site inspection, take photos, manually mark the installed area, and then manually input the project management system. This method is low in efficiency, high in labor intensity, and prone to problems such as input delay, omission and misfilling, causing the data to deviate from the actual situation. Even with the cooperation of BIM systems or progress management software, due to the lack of automatic acquisition means, these systems still remain at the level of “manual input—visual display”, and cannot realize the automatic update of the on-site state, the information is not timely, and it is difficult to support fine construction management.

[0004] The application of unmanned aerial vehicle aerial flight technology provides a new way for on-site data acquisition. The unmanned aerial vehicle can quickly cover the field area and obtain high-definition images, but it is currently mainly used for completion surveying and mapping or periodic inspection, and is difficult to support daily high-frequency monitoring. The existing technology generally relies on orthographic image stitching or three-dimensional reconstruction, which requires a large amount of computing power and time, and is not suitable for real-time updating on site. In addition, although some schemes combine YOLO and other recognition algorithms, they only focus on model detection accuracy, lack systematic processes such as task planning, image binding and statistical feedback, and therefore it is difficult to form a complete management system that can be implemented. SUMMARY

[0005] In view of the above problems, the application provides an intelligent photovoltaic construction progress management system and method based on aerial image analysis. The application integrates an edge computing terminal at the airport end of the unmanned aerial vehicle to realize local identification and statistics of aerial images. After the unmanned aerial vehicle performs a task, the image is directly transmitted to the terminal to complete target identification, and only the identification results (such as the number of components, completion status and confidence level) are uploaded, which significantly reduces the data volume and network dependence. The system automatically generates grid-level progress data, supports cumulative and incremental dual-perspective statistics, and automatically triggers a supplementary flight task according to the confidence level or abnormal state to form a closed loop of planning—acquisition—identification—feedback.

[0006] In order to achieve the above technical purposes, the present application provides a photovoltaic construction progress intelligent management and control method based on flight image analysis, which specifically comprises the following steps:

[0007] S1. Taking photovoltaic engineering design data and construction progress plan as basic data source, completing batch import and standardized modeling of data, arranging photovoltaic engineering design data and construction progress plan into unified basic data, and providing unified spatial reference and logical framework for subsequent flight planning, image recognition, grid matching and progress calculation;

[0008] S2. Based on the unified basic data in S1, the current construction state of each partition and component is displayed on the map in real time, the construction area to be monitored is automatically recognized, the ground coverage range of each flight point is calculated based on the path planning strategy of grid management, the flight path is generated, and the flight path is optimized to form a task planning result containing flight point sequence, path logic and execution parameters;

[0009] S3. The task planning result in S2 is packaged and parameter integrated to generate a flight task package, and the flight task package is pushed to the corresponding unmanned aerial vehicle airport end through the network, the unmanned aerial vehicle airport system automatically verifies the task integrity and detects the unmanned aerial vehicle state, and then executes the flight task to automatically complete the whole process of unmanned aerial vehicle take-off, task loading, route flight, photographing and return;

[0010] S4. The non-orthogonal grid matching method is used to establish the spatial correlation between the image file collected in the unmanned aerial vehicle flight process and the corresponding construction grid, and the spatial inverse calculation is directly performed by using the shooting parameters and flight attitude to realize the accurate binding of "one grid", and provide the spatial basis for subsequent identification and progress statistics; Figure One

[0011] S5. Based on the image-grid binding relationship in S4, the recognition algorithm module is called to detect and identify the number of photovoltaic components and supports in the image area in the grid, the identification result is generated as a structured data file and a labeled image file according to the grid number, and uploaded to the central server through the network to realize the real-time convergence of the on-site identification result;

[0012] S6. After the central server receives the uploaded identification result, the identification result is summarized and data fused according to the grid number, the cumulative and incremental progress proportion of each grid, partition and whole field range is calculated, the identification result is automatically compared with the construction plan data, the progress deviation analysis result is generated, and the progress deviation analysis result is visualized in two-dimensional or three-dimensional way.

[0013] ​A further technical solution of the present invention: When low recognition confidence, blurry image or abnormal recognition result is detected in step S6, a re-flight task list is automatically generated and sent to the UAV airport. In the next round of flight missions, abnormal grids are prioritized for re-shooting. After the re-shooting images are re-identified and uploaded, the statistical results are updated, forming a closed-loop processing flow of "collection - recognition - statistics - verification".

[0014] The preferred technical solution of the present invention is as follows: the photovoltaic engineering design data in step S1 mainly includes the following three types of information: equipment type information, construction coordinate information, and the zone number; the equipment type information includes the equipment number, specifications, and installation type of the photovoltaic bracket and photovoltaic module; the construction coordinate information includes the two-dimensional or three-dimensional coordinate data of each component in the design drawings, which is used to establish a spatial correspondence on the map; the zone number corresponds to a specific zone number for each component;

[0015] In step S1, the construction schedule partition number and component type are set at the smallest granularity. Input methods include file import and graphical interactive configuration. File import supports importing external Excel files. After the construction schedule data is imported in step S1, it is automatically matched with the design data to form a standardized multidimensional schedule table. This table is then displayed intuitively in a timeline view, allowing users to view the schedule status of each partition and component type on different dates. At any time node, the planned construction status of each grid unit or component is automatically calculated. These planned construction statuses will serve as the core basis for subsequent flight mission generation and image recognition progress comparison, providing a standardized reference for subsequent intelligent progress analysis and deviation detection.

[0016] In step S1, the photovoltaic engineering design data is imported in the form of spreadsheet files or database interfaces, and data consistency and logical verification are automatically performed. The logical verification mainly includes checking the uniqueness of the component number; verifying whether the coordinate field is complete and conforms to the coordinate system specification; checking whether the equipment type is in the equipment dictionary defined by the system; and detecting whether there are coordinate overlaps, duplicate numbers or missing items in the same partition.

[0017] A further technical solution of the present invention: In step S2, the status is distinguished by color marking. The construction status includes three states: not started, under construction, and completed. When the task is generated, the grid area with the status of under construction is automatically selected and used as the preliminary range of the flight area to be detected.

[0018] In step S2, the path planning strategy based on grid management is based on pre-defined grid management units, treating each grid as the smallest aerial photography task unit. Flight path planning is performed according to three principles: fixed-point shooting, vertical overhead shooting, and zero-overlap acquisition. The ground coverage area of ​​each waypoint is automatically calculated based on grid division information and camera parameters. The calculation logic is based on the UAV camera parameters: resolution W×H, focal length f, and sensor physical size. The unit is mm and the set flight altitude is h. The ground field of view of the camera is calculated in reverse. And, it is adapted to the grid edge length; the calculation formula is as follows:

[0019]

[0020]

[0021] , : Image resolution in the horizontal and vertical directions on the ground, m / pixel;

[0022] , : The actual width and length of the photo on the ground, in meters.

[0023] A further technical solution of the present invention: After waypoints are generated in step S2, flight routes are generated according to a predetermined scanning logic by manually setting a mode or automatically optimizing a mode. The automatic optimization mode automatically generates the shortest flight route based on the path optimization mechanism of spatial nearest neighbor search or principal component analysis algorithm. The generated waypoint and flight route information are finally integrated into a standard task file, which includes: waypoint coordinates, photo mode, gimbal attitude, flight speed, and take-off and landing point settings. The task package content in step S3 includes the flight route file, task parameters, and task identifier. The task parameters include flight altitude, speed, and photo interval. The task identifier includes partition number, grid list, and execution date.

[0024] A further technical solution of the present invention: The process of accurately binding the image files collected by the UAV during flight to the corresponding construction grid using a non-orthophoto mesh matching method in step S4 is as follows:

[0025] During flight, the drone transmits the captured aerial image files to the edge computing terminal at the drone's airport. The edge computing terminal parses the image's EXIF ​​metadata, extracting the latitude and longitude of the shooting center point, flight altitude h, heading angle θ, and camera parameters, including resolution. Focal length f, sensor physical dimensions Based on flight altitude and camera focal length, the ground coverage area is calculated. Assuming the UAV maintains a 90° vertical downward angle for shooting, the ground projection area of ​​the image is approximately rectangular. The system calculates the actual coverage size of this rectangle according to the calculation formula in step S2, based on flight altitude and camera parameters. , By combining the heading angle for spatial inverse calculation, the location range of the image in the geographic coordinate system is estimated. The image coverage area is then overlaid with the preset construction grid boundary to determine the target grid corresponding to the image, thus establishing a "one Figure One The image-grid binding relationship is established; the process of spatial inverse calculation based on the heading angle to estimate the location range of the image in the geographic coordinate system is as follows:

[0026] Considering the influence of the heading angle θ, the system further employs a two-dimensional rotation matrix to extend the center point of the image to the four corners:

[0027]

[0028] in, This is a two-dimensional planar rotation matrix, processed by coordinate rotation based on the drone's heading angle at the time of shooting; by combining the latitude and longitude (lng, lat) of the image center point with the corner offset ( , By combining calculations, the WGS84 coordinates of the four corner points of the image coverage area can be recovered, thus completing georeferenced processing.

[0029] A further technical solution of the present invention: In step S5, based on the image-mesh binding relationship in step S4, identification and statistics are performed through an edge computing terminal. The specific process is as follows:

[0030] S501. Perform resolution scaling, brightness correction, and distortion correction on the acquired image to ensure that the input image meets the model requirements;

[0031] The S502 edge computing terminal has a built-in lightweight deep learning detection model for identifying photovoltaic modules and supports. The model can simultaneously output the target category, bounding box position, and confidence value, enabling rapid determination of the construction status in the scene.

[0032] S503. Retain detection results with a confidence level higher than the set threshold, and count them according to the object type. The number of identified photovoltaic modules and supports will be bound to the corresponding grid ID to generate structured data records.

[0033] S504. After the recognition is completed, the edge computing terminal automatically outputs a structured recognition result file and annotated visualization image. The structured recognition result file includes the recorded grid number, component quantity, recognition time, and model confidence level. The annotated visualization image is generated by overlaying recognition boxes and category labels on the original aerial image to create an image file with detection labels for manual verification and subsequent auditing.

[0034] S505. After the recognition is completed, the edge computing terminal automatically performs confidence screening and quantity statistics on the recognition results, and uploads the recognition results to the central server in the form of structured files and compressed labeled images. Based on the recognition results, it automatically calculates the construction status and updates the progress database.

[0035] The preferred technical solution of this invention: In step S6, after the central server receives the recognition results and labeled images uploaded by each edge computing terminal, it automatically performs data parsing and aggregation; each recognition record contains core fields such as grid number, recognition time, number of supports, number of components, and confidence value. The central server uses the grid number as the primary key and automatically associates it with the existing plan table in the database to achieve data fusion; after completing the data fusion, it automatically performs two types of progress calculations:

[0036] Cumulative progress statistics: The total number of identified supports and components across the entire site up to the current moment is counted, and the ratio is calculated to the total design quantity to determine the overall completion rate.

[0037] Incremental progress statistics: By comparing the current identification results with the identification data of the previous period, the number of new installations is calculated, and a construction increment data table is generated;

[0038] The central server outputs statistical results at different granularities according to user needs, including statistics by region, by component type, and by time dimension. After the statistical results are generated, the actual completed data is automatically compared and analyzed with the planned construction data. The analysis includes plan deviation detection, time trend assessment, and critical path warning. Based on the progress analysis, it provides two- or three-dimensional visualization to intuitively display the construction completion status of each zone and the overall construction distribution information.

[0039] To achieve the above-mentioned technical objectives, the present invention also provides an intelligent control system for photovoltaic construction progress based on aerial image analysis. The system is used to execute the above-mentioned intelligent control method for photovoltaic construction progress based on aerial image analysis. The control system includes a central server, an unmanned aerial vehicle (UAV) airport system, and an edge computing terminal.

[0040] The central server includes a data source import and project modeling module, a task planning module, a task scheduling and image acquisition module, and a progress analysis and decision support module. The data source import and project modeling module uses photovoltaic engineering design data and construction schedule plans as the basic data source, completing batch data import and standardized modeling, and organizing the photovoltaic engineering design data and construction schedule plans into a unified basic database, providing unified basic data for subsequent flight planning, image recognition, grid matching, and progress calculation. The task planning module generates planning results including flight grids, waypoint sequences, and flight parameters based on the basic data provided by the data source import and project modeling module, and transmits these planning results as input data to the task scheduling and acquisition module. The task scheduling and image acquisition module encapsulates the planning results and integrates parameters to generate task packages, which are then distributed to the UAV airport system.

[0041] The unmanned aerial vehicle (UAV) airport system automatically verifies the integrity of the mission and detects the status of the UAV before executing the flight mission, automatically completing the entire process of UAV takeoff, mission loading, flight route, photography, and return.

[0042] The edge computing terminal includes a field perception and positioning module and an edge terminal image recognition and processing module. The field perception and positioning module uses a non-orthophoto grid matching method to establish a spatial association between image files collected during UAV flight and corresponding construction grids. It then directly performs spatial inverse calculations using shooting parameters and flight attitude to achieve "one-to-one" processing. Figure One The precise binding of the "grid" provides a spatial basis for subsequent identification and progress statistics; the edge terminal image recognition processing module detects and identifies the number of photovoltaic modules and brackets in the image area within the grid, and generates structured data files and labeled image files according to the grid number, and uploads them to the progress analysis and decision support module of the central server through the network;

[0043] The progress analysis and decision support module summarizes and merges data according to grid number based on the returned identification results, calculates the cumulative and incremental progress ratios of each grid, zone and the entire site, automatically compares the identification results with the construction plan data, generates progress deviation analysis results, and displays them in a two-dimensional or three-dimensional manner.

[0044] The preferred technical solution of the present invention is as follows: When the progress analysis and decision support module detects low recognition confidence, blurry images, or abnormal recognition results, it automatically generates a re-flight task list and sends it to the UAV airport system. In the next round of flight missions, it prioritizes re-shooting abnormal grids. After the re-shooting images are re-identified and uploaded, the statistical results are updated.

[0045] This invention employs a lightweight flight strategy using 90° overhead shooting, single-frame, non-orthophoto grid matching. It acquires images via fixed-point drone flight, with each image corresponding to a single grid cell, achieving "one..." Figure One The system uses spatial binding of "grids" and calculates the ground coverage area based on camera parameters (resolution, focal length, sensor size, flight altitude) to perform progress recognition without orthophoto conversion or stitching.

[0046] This invention combines photovoltaic design data, construction plans, and drone aerial imagery to achieve automatic identification, spatial positioning, and grid-based progress statistics for photovoltaic modules and supports. The system can cumulatively and incrementally calculate the construction completion status of each grid unit, ensuring the authenticity and timeliness of the progress data. Simultaneously, the system can generate visual progress reports, enabling low-intervention, intelligent construction management and effectively improving construction execution, management efficiency, and risk response capabilities. It solves the problems of data lag, large errors in manual recording, and lack of automated on-site sensing in existing photovoltaic construction progress management. Attached Figure Description

[0047] Figure 1 This is a flowchart of the method in this invention;

[0048] Figure 2 This is a system block diagram of the present invention;

[0049] Figure 3 This is a large-screen visualization effect diagram of the system in an embodiment of the present invention;

[0050] Figure 4 This is a visualization effect diagram of a two-dimensional map in an embodiment of the present invention;

[0051] Figure 5 This is a visualization effect of a three-dimensional map in an embodiment of the present invention. Specific Implementation

[0052] The present invention will be further described below with reference to embodiments. The technical solutions presented below are specific solutions of embodiments of the present invention and are not intended to limit the scope of the claimed invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention.

[0053] Example 1 provides an intelligent control system for photovoltaic construction progress based on aerial image analysis, such as... Figure 2As shown, a distributed network architecture of "central server - UAV airport system - edge computing terminal - UAV" is adopted. This architecture runs through the entire process of task planning, task packaging, issuance, execution, and identification, and is used to realize multi-level collaboration, hierarchical data processing, and real-time information interaction. The central server is responsible for task packaging, generation, issuance scheduling, status monitoring, and progress summarization; the UAV airport system is responsible for task scheduling, automatic take-off and landing, charging, and safety monitoring; the edge computing terminal is deployed at the airport and is used for task caching, image reception, and preliminary identification processing; the UAV terminal executes the actual flight and image acquisition tasks. In the architecture of this invention, the edge computing terminal is located on the UAV airport side and is a key node for task execution and data processing. Its core role is to realize a closed loop of task autonomy, on-site identification, and rapid data transmission. When the central server issues a flight task, the edge computing terminal automatically receives the task file and caches it locally to ensure that the task can still be executed independently in weak network or network outage environments.

[0054] The task distribution process of the photovoltaic construction progress intelligent management and control system in Example 1 is as follows: The server generates a task package based on the daily construction progress and flight area planning results, and the task package is pushed to the corresponding airport via the network; the corresponding UAV airport system automatically verifies the integrity of the task (number of waypoints, validity of coordinates, compliance of flight parameters); the UAV airport system detects the UAV status, including battery level, storage space, camera, gimbal angle, etc.; after the detection is passed, the task status changes to "pending execution" and waits for the takeoff window to be triggered. Once the server issues the execution command, the airport can automatically complete the entire process of UAV takeoff, task loading, flight route, photography, and return without manual intervention.

[0055] In Example 1, the central server includes a data source import and project modeling module, a task planning module, a task scheduling and image acquisition module, and a progress analysis and decision support module. The data source import and project modeling module uses photovoltaic engineering design data and construction schedule as the basic data source, completes batch import and standardized modeling of data, and organizes the photovoltaic engineering design data and construction schedule into a unified basic database, providing unified basic data for subsequent flight planning, image recognition, grid matching, and progress calculation. The task planning module generates planning results containing flight grids, waypoint sequences, and flight parameters based on the basic data provided by the data source import and project modeling module, and transmits the planning results as input data to the task scheduling and acquisition module. The task scheduling and image acquisition module encapsulates the planning results and integrates parameters to generate a task package, which is then sent to the UAV airport system.

[0056] The unmanned aerial vehicle (UAV) airport system automatically verifies the integrity of the mission and detects the status of the UAV before executing the flight mission, automatically completing the entire process of UAV takeoff, mission loading, flight route, photography, and return.

[0057] The edge computing terminal includes a field perception and positioning module and an edge terminal image recognition and processing module. The field perception and positioning module uses a non-orthophoto grid matching method to establish a spatial association between image files collected during UAV flight and corresponding construction grids. It then directly performs spatial inverse calculations using shooting parameters and flight attitude to achieve "one-to-one" processing. Figure One The precise binding of the "grid" provides a spatial basis for subsequent identification and progress statistics; the edge terminal image recognition processing module detects and identifies the number of photovoltaic modules and brackets in the image area within the grid, and generates structured data files and labeled image files according to the grid number, and uploads them to the progress analysis and decision support module of the central server through the network;

[0058] The progress analysis and decision support module summarizes and merges data according to grid number based on the returned identification results, calculates the cumulative and incremental progress ratios of each grid, zone and the entire site, automatically compares the identification results with the construction plan data, generates progress deviation analysis results, and displays them in a two-dimensional or three-dimensional manner.

[0059] When the progress analysis and decision support module detects low recognition confidence, blurry images, or abnormal recognition results, it automatically generates a re-flight task list and sends it to the UAV airport system. In the next round of flight missions, it prioritizes re-shooting abnormal grids. After the re-shooting images are re-recognized and uploaded, the statistical results are updated.

[0060] This invention integrates an edge computing terminal at the UAV airport end to achieve localized recognition and statistics of aerial images. After the UAV performs a mission, the images are directly transmitted to the terminal to complete target recognition. Only the recognition results (such as the number of components, completion status, confidence level, etc.) are uploaded, which significantly reduces the amount of data and network dependence. The system automatically generates grid-level progress data, supports cumulative and incremental dual-caliber statistics, and automatically triggers re-flight tasks based on confidence level or abnormal status, forming a closed loop of planning-collection-recognition-feedback.

[0061] Example 2 provides a method for intelligent control of photovoltaic construction progress based on aerial image analysis. This method is implemented based on the intelligent control system for photovoltaic construction progress based on aerial image analysis in Example 1. The implementation process is as follows: Figure 1 As shown, the specific steps are as follows:

[0062] S1. The photovoltaic construction progress intelligent management and control system (hereinafter referred to as the "management and control system") in the embodiment uses photovoltaic engineering design data and construction progress plan as the basic data source in the early stage of system deployment, completes the batch import and standardized modeling of data, and organizes the photovoltaic engineering design data and construction progress plan into unified basic data; this step is the basic link of the entire system operation, providing a unified spatial benchmark and logical framework for subsequent flight planning, image recognition, grid matching and progress calculation.

[0063] The system first imports the basic design data of the photovoltaic project, which mainly includes the following three types of key information:

[0064] (1) Equipment type information: Record the equipment number, specifications, installation type and other attributes of the main facilities such as photovoltaic brackets and photovoltaic modules.

[0065] (2) Construction coordinate information: Read the two-dimensional or three-dimensional coordinate data of each component in the design drawings (such as pile foundation location, support layout location, etc.) to establish spatial correspondence in the map.

[0066] (3) Partition number: Each component corresponds to a specific partition number (such as A area, B area, etc.) to facilitate subsequent progress control.

[0067] The system supports importing design data in spreadsheet file format (such as .CSV, .XLSX) or database interface format, and automatically performs data consistency and logical checks, mainly including: checking the uniqueness of component numbers; verifying whether coordinate fields are complete and conform to coordinate system specifications (such as WGS84 or local projection system); checking whether the equipment type is in the system-defined equipment dictionary; detecting whether there are coordinate overlaps, duplicate numbers, or missing items in the same partition; through the above verification mechanism, the system can discover potential data logic errors during the import stage, effectively avoiding problems such as positioning deviation and recognition failure in the subsequent flight matching and recognition stages.

[0068] After the design data is imported, the system imports or configures the construction schedule data. The construction schedule is set at the smallest granularity of "zone number + component type," eliminating the need for individual component configuration and significantly reducing manual data entry. The system supports two schedule input methods:

[0069] File import method: Supports importing external Excel files. Data fields should include "partition number", "component type", "planned start time", "planned completion time", etc.

[0070] Graphical interactive configuration method: Users can visually set or batch adjust the planned time through the system interface, and selective modification and time synchronization are supported.

[0071] After the planned data is imported, the system automatically matches it with the design data to form a standardized multi-dimensional plan table (partition-type-time). The system displays this table intuitively in a timeline view, allowing users to view the planned status of each partition and type of component on different dates. At any given time point, the system can automatically calculate the planned construction status ("not started," "under construction," or "completed") for each grid unit or component. These planned statuses will serve as the core basis for subsequent flight mission generation and image recognition progress comparison, providing a standardized reference for subsequent intelligent progress analysis and deviation detection.

[0072] S2. After importing the design data and construction plan, the intelligent photovoltaic construction progress control system of this invention enters the flight mission planning stage. The goal of this stage is to automatically identify the construction areas to be monitored based on the current construction progress status, and generate the optimal flight path and mission file to achieve automated and standardized drone aerial photography mission scheduling. Specifically, based on the unified basic data from step S1, the system displays the current construction status of each zone and component on a map in real time, automatically identifies the construction areas to be monitored, calculates the ground coverage of each waypoint based on a grid-based path planning strategy, generates flight paths, and optimizes the flight paths to form a mission planning result containing waypoint sequences, path logic, and execution parameters.

[0073] The system uses color coding to distinguish statuses: red indicates "not started," yellow indicates "under construction," and green indicates "completed." When a daily task is generated, the system automatically filters out the grid areas with the status "under construction" and uses them as the initial range for the flight area to be monitored.

[0074] To achieve the goals of accurate data collection, efficient operation, and lightweight processing, this invention adopts a path planning strategy based on grid management. During the task generation phase, the platform uses pre-defined grid management units as a basis, treating each grid as the smallest aerial photography task unit, and plans the flight path according to the following three principles:

[0075] Fixed-point shooting principle: The center point or geometric center of each grid is set as the target point for aerial photography, eliminating the need for continuous flight strips or high-overlap shooting, ensuring "one point, one image".

[0076] Vertical overhead shooting principle: The drone gimbal is locked at a 90° overhead angle (vertically downward) to avoid perspective distortion and geometric shift, and to ensure that the image size can be accurately calculated.

[0077] Zero-overlap acquisition principle: The coverage areas of adjacent grid waypoints do not overlap, reducing redundant shooting and data duplication, and reducing storage and processing burden.

[0078] Based on grid division information and camera parameters (resolution, focal length, sensor size, flight altitude), the platform automatically calculates the ground coverage area for each waypoint and generates a flight path accordingly. Its core calculation logic is as follows:

[0079] To ensure that the image coverage area matches the grid size, the system uses the UAV camera parameters: resolution W×H, focal length f, and sensor physical size. (Unit: mm) and set flight altitude h, to calculate the camera's ground field of view. And, it is adapted to the grid edge length; the calculation formula is as follows:

[0080]

[0081]

[0082]

[0083] In the formula: , : Horizontal and vertical resolution of the image on the ground (m / pixel);

[0084] , : Actual width and length of the photo's coverage on the ground (unit: meters);

[0085] The system compares the grid side length with the ground coverage area to ensure that each image completely covers a grid cell without crossing the boundary; by automatically adjusting flight altitude or camera focal length parameters, the image ground size can be strictly matched with the grid boundary, achieving "one-to-one" accuracy. Figure One Spatial binding of "grid".

[0086] After waypoints are generated, the system determines the flight order according to a predetermined scanning logic. The system supports two path sorting modes:

[0087] Manual setting mode: Users can specify the scanning direction (such as from left to right, from top to bottom), which is used for regular sites or small areas;

[0088] Automatic optimization mode: The system adopts a path optimization mechanism based on the Nearest Neighbor search or Principal Component Analysis (PCA) algorithm to automatically generate the shortest route and reduce unnecessary turns and energy consumption.

[0089] The generated waypoint and route information is ultimately integrated into a standard mission file (such as KML or DJI Waypoint format), which includes: waypoint coordinates (latitude, longitude, and altitude); shooting mode (fixed-point single shot or timed shooting); gimbal attitude (90° overhead fixed shot); and flight speed and takeoff / landing point settings. The system automatically adds a mission number, generation time, and area identification information to the mission file for easy airport-side scheduling and management.

[0090] After path generation and optimization, the platform only generates a "task planning result," which includes waypoint sequence, path logic, and execution parameters, but has not yet formed a formal execution task package. This planning result will be transmitted as input data to the central server, which will then generate the formal task package in subsequent stages and distribute it to the airport for execution.

[0091] S3. After completing the flight area planning and path generation, the intelligent photovoltaic construction progress control system enters the task issuance and airport execution phase. In this phase, the system first encapsulates and integrates the flight planning results output from the previous phase through the central server, generating a formal task package; then, it completes the issuance of the task package, airport-side execution, and task status feedback. The central server pushes the flight task package to the corresponding UAV airport via the network. The UAV airport system automatically verifies the task integrity, detects the UAV status, and then executes the flight task, automatically completing the entire process of UAV takeoff, task loading, flight path, photography, and return.

[0092] The formal mission package includes: flight path file; mission parameters (altitude, speed, photo interval); mission identifiers (zone number, grid list, execution date). The mission package is distributed to the corresponding UAV airport via the network. After receiving it, the airport system automatically verifies the mission's integrity and prepares accordingly: verifying waypoint coordinates and safe zone; checking the aircraft's battery level and camera status; loading the mission and waiting to execute it.

[0093] Through the above process, the system realizes the transformation from route planning results to executable task packages, laying the foundation for subsequent task scheduling, UAV execution, and progress recognition.

[0094] S4. After the UAV completes its flight, the captured image files are transmitted back to the edge computing terminal in real time. The edge computing terminal uses a non-orthophoto mesh matching method to establish a spatial association between the image files captured during the UAV's flight and the corresponding construction mesh. It then directly performs spatial inverse calculations using the shooting parameters and flight attitude, achieving "one-to-one" spatial mapping. Figure OneThe precise binding of the "grid" provides a spatial foundation for subsequent identification and progress statistics; specifically as follows: After the UAV completes its flight mission, the collected aerial image files are transmitted to the edge computing terminal at the UAV airport. The edge computing terminal parses the image EXIF ​​metadata and extracts the latitude and longitude (longitude lng, latitude la) of the shooting center point, the flight altitude h, the heading angle θ, and the camera parameters (resolution). Focal length f, sensor physical dimensions (Unit: mm). By reading these parameters, the system can accurately obtain the spatial positioning and imaging geometric information of each photo, providing input for subsequent calculation of the ground projection range; the ground coverage range is calculated based on the flight altitude and camera focal length, and spatial inverse calculation is performed in combination with the heading angle to estimate the position range of the image in the geographic coordinate system, realizing geographic reference registration under non-orthophoto conditions.

[0095] Assuming the drone maintains a 90° vertical overhead view, the ground projection area of ​​the image is approximately rectangular. The platform calculates the actual coverage size of this rectangle based on the flight altitude and camera parameters. , (The specific calculation formula is in step S2). Considering the influence of the heading angle θ, the platform further uses a two-dimensional rotation matrix to extend the center point of the photo to the four corners:

[0096]

[0097] in, This is a two-dimensional planar rotation matrix, processed by rotating the coordinates based on the drone's heading angle at the time of capture. This is achieved by rotating the image center point's latitude and longitude (lng, lat) along with the corner offset (...). , By combining calculations, the WGS84 coordinates of the four corner points of the image coverage area can be recovered, thus completing georeferenced processing.

[0098] After obtaining the image coverage area, the system automatically determines whether the center point coordinates of the photo are within the boundary of a certain grid. If the condition is met, the photo is considered to correspond one-to-one with that grid, and a binding relationship is established. To ensure binding accuracy, the system executes the following logic:

[0099] If multiple photos exist within the same grid, the system prioritizes them based on aerial photography time, shooting angle, and GSD accuracy, and selects the best image as the main image.

[0100] If the grid boundary is covered by multiple overlapping maps, the system will primarily use the principle of center point attribution to avoid cross-grid statistics.

[0101] The system automatically generates a unique "grid ID + timestamp" identifier for each image, which is used for subsequent matching and recognition result tracking.

[0102] After binding is completed, the platform overlays the corresponding grid boundary contour on the image, retaining only the area as the effective recognition area. The parts of the image that exceed the grid will be automatically masked or cropped to ensure that the recognition results are strictly limited to the specified construction unit.

[0103] S5. Based on step S4, the edge computing terminal, according to the image-grid binding relationship, locally calls the recognition algorithm module to detect and count the number of photovoltaic modules and supports in the image area within the grid. The recognition results generate structured data files and labeled image files according to the grid number, and are uploaded to the central server via the network to achieve real-time aggregation of on-site recognition results. At the same time, by completing the parsing and recognition at the edge, the load on the network and server from large-scale image backhaul is reduced. The specific process is as follows:

[0104] The system automatically categorizes and archives data according to grid number, timestamp, and task batch. Then, the edge computing terminal calls its local recognition algorithm module to perform target detection and progress recognition on the images, generating two types of output results: one is a structured recognition result (JSON or CSV format), including the number of components in each grid, completion status, and recognition confidence level; the other is a compressed image file after recognition, with detection boxes, category labels, and confidence level markers overlaid on the original photo to form a visual annotation map. The edge computing terminal packages the recognition results and annotated images together, compresses and encrypts them, and then uploads them to the server for subsequent progress summarization and verification display.

[0105] To ensure the safety and stability of unmanned flight missions, the UAV airport system of this invention is equipped with a comprehensive automatic detection and safety control mechanism. Before mission execution, the airport system automatically performs battery level checks, propeller and gimbal self-checks, GPS positioning signal verification, and environmental wind speed assessment to ensure flight conditions meet safety standards. During flight, the system monitors the UAV's attitude angle, flight speed, heading deviation, and communication signal strength in real time. If any anomaly is detected (such as signal loss, excessive wind speed, insufficient battery power, or flight deviation), an automatic return-to-home or interruption protection procedure is immediately executed to ensure equipment and data security. After the mission, the UAV automatically lands on the parking apron and connects to the charging module. The airport system then completes mission status feedback and equipment reset. The central server can view mission progress and equipment status in real time, recording status changes such as "pending execution—in execution—completed—abnormal termination," and supports automatic rescheduling and re-flight execution of abnormal missions. Through these multi-layered monitoring and safety mechanisms, the system ensures the controllability, reliability, and traceability of flight mission operations while guaranteeing unmanned operation throughout the entire process.

[0106] S5. Based on the image-grid binding relationship in step S4, the edge computing terminal calls the recognition algorithm module to detect and count the number of photovoltaic modules and supports in the image area within the grid. The recognition results generate structured data files and labeled image files according to the grid number, and are uploaded to the central server via the network to achieve real-time aggregation of on-site recognition results. This stage is the key link in realizing on-site intelligent analysis and rapid generation of progress data in this invention. By deploying an edge computing terminal at the UAV airport, the system can immediately perform image recognition, status determination, and result generation after data acquisition, realizing a "collection-as-analysis" closed-loop processing mode on-site. The specific process is as follows:

[0107] S501. Data Loading and Preprocessing: The acquired images are scaled for resolution, luminance corrected, and distortion corrected to ensure the input images meet the model requirements. The platform automatically extracts the effective region corresponding to the target mesh and filters out parts exceeding the boundary.

[0108] S502. Model Inference and Object Detection: The edge computing terminal has a built-in lightweight deep learning detection model (such as YOLOv8 or equivalent structure) to identify construction objects such as photovoltaic modules and brackets; the model can simultaneously output the target category, bounding box position and confidence value to achieve rapid determination of the construction status in the scene.

[0109] S503. Result Filtering and Statistical Summary: The system only retains detection results with a recognition confidence level higher than a set threshold (e.g., 0.6) and performs counting statistics based on the object type; the number of identified components and supports will be bound to the corresponding grid ID to generate structured data records.

[0110] S504. Generation of Result Files and Annotated Images: After recognition is completed, the terminal automatically outputs two types of result files; Structured recognition result file (JSON / CSV format): records fields such as grid number, number of components, recognition time, and model confidence level; Annotated visualization image (stored after compression): the recognition box and category label are superimposed on the original aerial image to generate an image file with detection labels for manual verification and subsequent auditing.

[0111] After the identification is completed, the edge computing terminal automatically performs confidence screening and quantity statistics on the identification results, and uploads the results to the central server in the form of structured files and compressed labeled images. The system automatically calculates the construction status and updates the progress database in the background based on the identification results.

[0112] S6. After receiving the uploaded identification results, the central server summarizes and merges the data according to the grid number, calculates the cumulative and incremental progress ratios for each grid, zone, and the entire site, and automatically compares the identification results with the construction plan data to generate progress deviation analysis results (such as...). Figure 3 As shown), and in two dimensions (such as...)Figure 4 (as shown) or three-dimensional (such as) Figure 5 The data is visualized in the manner shown (as indicated). The core objective of this stage is to compare the identified actual construction data with the pre-imported design plan data in the platform to generate multi-dimensional construction progress statistics at the grid, zone, and overall site levels, thereby achieving comprehensive quantification and dynamic analysis of the construction status. The specific process is as follows:

[0113] (1) Data aggregation and structured storage: After the server receives the recognition results and labeled images uploaded by each edge computing terminal, the system automatically performs data parsing and aggregation; each recognition record contains core fields such as grid number, recognition time, number of supports, number of components, and confidence value. The system uses the grid number as the primary key and automatically associates it with the existing plan table in the database to achieve data fusion. For multiple recognition results of the same grid, the system sorts them according to the timestamp, retains only the most recent recognition data as the current state, and saves historical records for trend analysis. Through this mechanism, the platform can form a complete construction data time series, providing a data foundation for subsequent progress change trends and deviation judgment.

[0114] (2) Cumulative progress and incremental progress calculation: After the data fusion is completed, the system automatically performs two types of progress calculations:

[0115] Cumulative progress statistics: The total number of identified supports and components across the entire site up to the current moment is counted, and the ratio is calculated to the total design quantity to determine the overall completion rate.

[0116] Incremental progress statistics: By comparing the current identification results with the identification data of the previous cycle (such as the previous flight mission), the number of new installations is calculated, and a construction increment data table is formed.

[0117] The system can output statistical results at different granularities according to user needs, including: statistics by region (partition number); statistics by component type (support, component); and statistics by time dimension (day, week, month).

[0118] These statistics are not only used to generate daily and weekly reports, but also provide management with a basis for dynamic construction efficiency assessment.

[0119] (3) Comparison and analysis between plan and actual results: After the statistical results are generated, the system will automatically compare and analyze the actual completed data with the planned construction data. This analysis includes:

[0120] Plan Deviation Detection: Calculate the difference between the "planned completion rate" and the "actual completion rate" for each partition or grid. If the deviation exceeds the set threshold (e.g., ±10%), the system will automatically mark it as "lagging behind" or "ahead of schedule".

[0121] Time trend assessment: Generate progress change curves based on multiple identification records to analyze construction speed trends;

[0122] Critical path warning: Based on the task plan, identify nodes in critical partitions that are at risk of delay and automatically push warning information.

[0123] Through these analytical results, managers can intuitively grasp the overall construction status, adjust resource allocation and work plans in a timely manner, and achieve scientific management and dynamic control.

[0124] (4) Visualization and Report Output; Based on progress analysis, the system provides various visualization formats such as two-dimensional and three-dimensional (e.g., Figures 3 to 5 As shown in the image, the system can intuitively display the construction completion status of each zone and the overall construction distribution information. Simultaneously, the system can automatically generate standardized progress reports, including task execution batches, identified data summary tables, deviation analysis results, and graphical displays. The reports can be exported to PDF or Excel format for use in project meetings, owner presentations, and as-built documentation archiving.

[0125] S7. To ensure data accuracy and complete on-site coverage, the platform features anomaly detection and automatic go-around functionality. When the following conditions occur in the identification results, the system will automatically mark them as abnormal grids: identification confidence level below a set threshold; image blurry or severely occluded; abnormally reduced number of identifications for the same grid (progress rollback). For abnormal grids, the system automatically generates a go-around task list and pushes it to the corresponding airport, allowing the drone to prioritize re-shooting in the next task window, thus achieving automatic correction and dynamic improvement of progress data. The re-shot images are re-identified and uploaded, updating the statistical results and forming a closed-loop processing flow of "acquisition—identification—statistics—verification".

[0126] This invention addresses the problems of data lag, large errors in manual recording, and lack of automated on-site sensing in photovoltaic (PV) construction progress management. It proposes an intelligent control method and system based on aerial image analysis. By combining PV design data, construction plans, and UAV aerial images, the system achieves automatic identification, spatial positioning, and grid-based progress statistics for PV modules and supports. The system can cumulatively and incrementally calculate the construction completion status of each grid unit, ensuring the authenticity and timeliness of progress data. Simultaneously, the platform can generate visual progress reports, enabling low-intervention, intelligent construction management and effectively improving construction execution, management efficiency, and risk response capabilities.

[0127] The above description is merely one embodiment of the present invention, and while it is detailed and specific, it should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. An intelligent management and control method for photovoltaic construction progress based on flight image analysis, characterized in that, Specifically comprising the following steps: S1. Taking photovoltaic engineering design data and construction progress plan as the basis data source, completing batch import and standardized modeling of data, and arranging the photovoltaic engineering design data and construction progress plan into unified basis data, thereby providing a unified spatial reference and logical framework for subsequent flight planning, image recognition, grid matching and progress calculation; S2. Based on the unified basis data in S1, displaying the current construction state of each subzone and each component on the map in real time, automatically identifying the construction area to be monitored, calculating the ground coverage range of each waypoint based on the path planning strategy of grid management, generating a flight path, and optimizing the flight path to form a task planning result containing a waypoint sequence, path logic and execution parameters; S3. Packaging and parameter integration of the task planning result in S2 to generate a flight task package, and pushing the flight task package to the corresponding unmanned aerial vehicle airport end through the network, and the unmanned aerial vehicle airport system automatically verifies the task integrity and detects the unmanned aerial vehicle state, and then executes the flight task, automatically completes the whole process of unmanned aerial vehicle take-off, task loading, route flight, photographing and return; S4. Using a non-orthogonal grid matching method to establish a spatial correlation between the image files collected during the unmanned aerial vehicle flight and the corresponding construction grid, and directly performing spatial inverse calculation using the shooting parameters and flight attitude to achieve precise binding of "one image and one grid", thereby providing a spatial basis for subsequent identification and progress statistics; S5. Based on the image-grid binding relationship in S4, calling the recognition algorithm module to detect and identify the number of photovoltaic components and supports in the image area in the grid, and generating a structured data file and a labeled image file according to the grid number, and uploading them to the central server through the network to realize real-time convergence of the on-site identification result; S6. After receiving the uploaded identification result, the central server summarizes and fuses the data according to the grid number, calculates the cumulative and incremental progress proportion of each grid, subzone and the whole field, automatically compares the identification result with the construction plan data, generates a progress deviation analysis result, and visualizes it in two or three dimensions. 2.The photovoltaic construction progress intelligent management and control method based on flight image analysis according to claim 1, characterized in that: When low recognition confidence, blurred images or abnormal identification results are detected in S6, a reflight task list is automatically generated and sent to the unmanned aerial vehicle airport, and the abnormal grid is preferentially rephotographed in the next round of flight task. After the rephotographed images are reidentified and uploaded, the statistical result is updated, forming a closed-loop processing flow of "collection-identification-statistics-recheck". 3.The photovoltaic construction progress intelligent management and control method based on flight image analysis according to claim 1 or 2, characterized in that: The photovoltaic engineering design data in S1 mainly includes the following three types of information: equipment type information, construction coordinate information and subzone number; the equipment type information includes the equipment number, specification and model, and installation type of photovoltaic supports and photovoltaic components; the construction coordinate information includes two-dimensional or three-dimensional coordinate data of each component in the design drawing, which is used to establish a spatial correspondence in the map; The subzone number corresponds to a specific subzone number for each component. The construction progress plan partition number + component type in the S1 step is set as the minimum granularity, and the input mode includes a file import mode and a graphical interactive configuration mode, and the file import supports importing external Excel files; after the construction progress plan data is imported in the S1 step, the data is automatically matched with design data to form a standardized multi-dimensional plan table, and the table is intuitively displayed in a time axis view, so that a user can view the plan state of each partition and each type of component on different dates; at any time node, the planned construction state of each grid unit or component is automatically calculated, and these planned construction states serve as a core basis for subsequent flight task generation and image recognition progress comparison, and provide a standardized comparison for subsequent intelligent progress analysis and deviation detection; The photovoltaic engineering design data in the S1 step is imported in the form of a spreadsheet file or a database interface, and data consistency and logical verification are automatically performed, and the logical verification mainly includes checking the uniqueness of the component number; verifying whether the coordinate field is complete and conforms to the coordinate system specification; correcting whether the equipment type is in the equipment dictionary defined by the system; detecting whether there is coordinate overlap, number duplication or missing items in the same partition. 4.The photovoltaic construction progress intelligent management and control method based on flight image analysis according to claim 1 or 2, characterized in that: In the S2 step, the states are distinguished by color identification, and the construction states include three states of not started, in construction and completed, the grid areas in the state of in construction are automatically selected when the task is generated, and the grid areas are used as the preliminary range of the flight areas to be detected; In the S2 step, the path planning strategy based on grid management takes the pre-defined grid management unit as the basis, regards each grid as the smallest aerial photography task unit, and plans the flight route according to three principles of fixed-point shooting, vertical shooting and zero-overlap collection; According to the grid division information and the camera parameters, the ground coverage range of each flight point is automatically calculated. The calculation logic is based on the unmanned aerial vehicle camera parameters: resolution WxH, focal length f, sensor physical size , unit mm and set flight height h, inverse calculation of camera ground field of view range ( ), and adapt to the grid length; the calculation formula is as follows: ； ； ； , : horizontal and vertical resolution of the image on the ground, m / pixel; , : the actual width and length of the photo on the ground, in m.

5. The photovoltaic construction progress intelligent management and control method based on flight image analysis according to claim 1 or 2, characterized in that: After the flight points are generated in the S2 step, the flight order is determined according to the established scanning logic by manually setting mode or automatic optimization mode, the automatic optimization mode generates the shortest flight route based on the path optimization mechanism of the spatial nearest neighbor search or principal component analysis algorithm, and the generated flight points and flight route information are finally integrated into a standard task file, which includes the flight point coordinates, shooting mode, gimbal attitude and flight speed and take-off and landing point setting; the task package content in the S3 step includes the flight route file and task parameters and task identification, the task parameters include the flight height, speed and shooting interval, and the task identification includes the partition number, grid list and execution date. 6.The photovoltaic construction progress intelligent management and control method based on flight image analysis of claim 4, characterized in that: In the S4 step, the non-orthogonal grid matching method is used to accurately bind the image files collected in the flight process of the unmanned aerial vehicle and the corresponding construction grid, and the process is as follows: In the flight process of the unmanned aerial vehicle, the aerial image file collected is transmitted to the edge computing terminal of the unmanned aerial vehicle airport end, the edge computing terminal analyzes the image EXIF metadata, extracts the shooting center point longitude and latitude, flight height h, heading angle θ and camera parameters, the camera parameters include resolution , focal length f, sensor physical size , the ground coverage range is calculated based on the flight height and the camera focal length, under the premise that the unmanned aerial vehicle maintains 90° vertical shooting, the ground projection area of the image is approximately rectangular, the system calculates the actual coverage size of the rectangle according to the calculation formula in step S2 、 , and combines the heading angle to perform spatial inverse calculation to calculate the position range of the image in the geographic coordinate system, superimposes the image coverage range and the preset construction grid boundary to determine the target grid corresponding to the image, and establishes the image-grid binding relationship of "one image one grid"; the process of spatial inverse calculation combined with the heading angle to calculate the position range of the image in the geographic coordinate system is as follows: Considering the influence of the heading angle θ, the system further expands the center point of the photo to the four corners by using a two-dimensional rotation matrix: ； in, This is a two-dimensional planar rotation matrix, processed by coordinate rotation based on the drone's heading angle at the time of shooting; by combining the latitude and longitude (lng, lat) of the image center point with the corner offset ( , By combining calculations, the WGS84 coordinates of the four corner points of the image coverage area can be recovered, thus completing georeferenced processing.

7. The photovoltaic construction progress intelligent management and control method based on flight image analysis according to claim 6, characterized in that: In the S5 step, based on the image-grid binding relationship in the S4 step, the identification and statistics are performed by the edge computing terminal, and the specific process is as follows: In the S501, the resolution of the collected image is scaled, the brightness is corrected, and the distortion is corrected to ensure that the input image meets the model requirements; In the S502, the edge computing terminal is built-in with a lightweight deep learning detection model for identifying photovoltaic components and supports, and the model can simultaneously output target categories, bounding box positions and confidence values, so as to realize rapid determination of the construction state in the scene; S503, retain the detection results with recognition confidence higher than the set threshold, and count according to the object type, the number of recognized photovoltaic components and supports will be bound to the corresponding grid ID, and structured data records are generated; S504, after the identification is completed, the edge computing terminal automatically outputs the structured identification result file and the labeled visual image, the structured identification result file includes the record grid number, the component number, the identification time and the model confidence; the labeled visual image is the identification frame and the category label superimposed on the original aerial photograph, and an image file with detection mark is generated, which is used for manual checking and subsequent audit; S505, after the identification is completed, the edge computing terminal automatically performs confidence screening and quantity statistics on the identification result, and uploads the identification result to the central server in the form of structured file and compressed labeled image, and automatically calculates the construction state update progress database according to the identification result. 8.The photovoltaic construction progress intelligent management and control method based on flight image analysis of claim 4, characterized in that: S6 After the central server receives the identification result and labeled image uploaded by each edge computing terminal, data analysis and summarization are automatically performed; each identification record contains grid number, identification time, support number, component number, confidence value and other core fields, the central server takes grid number as the primary key, automatically associates with the existing plan table in the database, realizes data fusion; after completing data fusion, two types of progress calculation are automatically performed: Cumulative progress statistics: the sum of the identified support and component numbers in the whole field range is calculated, and the ratio with the total design quantity is calculated to obtain the overall completion ratio; Incremental progress statistics: by comparing the current identification result with the identification data of the last period, the newly installed quantity is calculated to form the construction incremental data table; The central server outputs statistical results of different granularities according to user demand, including statistics according to area, statistics according to component type and statistics according to time dimension, and after the statistical results are generated, the actual completion data and the planned construction data are automatically compared and analyzed, the analysis includes plan deviation detection, time trend evaluation and key path warning, on the basis of progress analysis, two or three-dimensional visual display forms are provided to intuitively display the construction completion state of each partition and the whole field construction distribution information.

9. An intelligent photovoltaic construction progress management and control system based on flight image analysis, characterized in that: The system is used to execute the photovoltaic construction progress intelligent control method based on aerial image analysis according to any one of claims 1 to 8, and the control system comprises a central server, a UAV airport system and an edge computing terminal; The central server comprises a data source import and project modeling module, a task planning module, a task scheduling and image acquisition module and a progress analysis and decision support module; The data source import and project modeling module takes the photovoltaic engineering design data and the construction progress plan as the basic data source, completes batch data import and standardized modeling, arranges the photovoltaic engineering design data and the construction progress plan into a unified basic database, and provides unified basic data for subsequent aerial planning, image recognition, grid matching and progress calculation; The task planning module generates a planning result containing a flight grid, a waypoint sequence and flight parameters according to the basic data provided by the data source import and project modeling module, and transmits the planning result as input data to the task scheduling and image acquisition module; The task scheduling and image acquisition module encapsulates and integrates the planning result to generate a task package, and issues the task package to the UAV airport system; The UAV airport system automatically checks the task integrity and detects the UAV state, and then executes the flight task to automatically complete the whole process of UAV take-off, task loading, route flight, photographing and return; The edge computing terminal includes a field perception and positioning module and an edge terminal image recognition processing module; The field perception and positioning module uses a non-orthogonal grid matching method to establish a spatial correlation between the image files collected during the UAV flight and the corresponding construction grid, directly performs spatial inverse calculation using the shooting parameters and flight attitude, realizes precise binding of "one image and one grid", and provides a spatial basis for subsequent identification and progress statistics; the edge terminal image recognition processing module detects and identifies the number of photovoltaic modules and supports in the grid image area, generates a structured data file and a labeled image file according to the grid number, and uploads the identification result to the progress analysis and decision support module of the central server through the network; The progress analysis and decision support module summarizes and data fuses the identification results according to the grid number, calculates the cumulative and incremental progress proportion of each grid, sub-area and the whole field, automatically compares the identification results with the construction plan data, generates a progress deviation analysis result, and visually displays the result in two or three dimensions. 10.The photovoltaic construction progress intelligent management and control system based on flight image analysis of claim 9, characterized in that: When the progress analysis and decision support module detects low recognition confidence, image blur or abnormal identification result, it automatically generates a reflight task list and issues the list to the UAV airport system, which prioritizes abnormal grids for rephotographing in the next round of flight tasks, and updates the statistical results after the rephotographed images are reidentified and uploaded.

Images