Satellite and unmanned aerial vehicle cooperative large-scale photovoltaic power station construction progress investigation method

Abstract

The application discloses a satellite and unmanned aerial vehicle cooperative large-range photovoltaic power station construction progress investigation method, and belongs to the technical field of photovoltaic power station data processing. The technical problem to be solved is that the existing investigation method has low operation efficiency for full unmanned aerial vehicle coverage of a large-range photovoltaic power station. The method comprises the following steps: S1, remote sensing data set retrieval is performed according to set conditions of the photovoltaic power station; S2, multi-band image data is fused to generate image data; S3, photovoltaic panel targets in the image are extracted to generate photovoltaic panel laying areas; S4, a latest construction area is calculated as a construction target area; S5, high-precision unmanned aerial vehicle image data of the construction target area is collected; S6, a high-precision orthographic image is generated, and component-level targets in the high-precision orthographic image are extracted; S7, the number of the component-level targets is updated; S8, statistical analysis is performed according to the number of completed constructions and planned design component-level targets, and construction progress is obtained.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power plant data processing technology, specifically involving a method for large-scale photovoltaic power plant construction progress survey using satellite and drone collaboration. Background Technology

[0002] Construction progress monitoring is a core aspect of project management, crucial for ensuring on-time project delivery, cost control, and optimized resource allocation. With the development of remote sensing, computer vision, and drone technologies, automated progress monitoring has become a research hotspot. Existing technologies have made significant progress by extracting the quantity and spatial distribution of targets such as photovoltaic panels, supports, and piles in photovoltaic power plant sites using drone imagery, and then comparing this data with construction design drawings to assess construction progress.

[0003] However, relying solely on drones to collect images of photovoltaic power plant sites presents significant limitations when dealing with large-scale, distributed, and highly repetitive photovoltaic power plant construction projects. Large-scale photovoltaic power plants typically cover tens of square kilometers and are often located in remote areas such as deserts and Gobi, where the environment is harsh. In this context, existing methods for construction progress surveys mainly face the following technical challenges: (1) Low efficiency and high cost of investigation: Existing UAV inspection solutions, whether used for operation and maintenance or construction monitoring, usually adopt a "full coverage" or "carpet-style" flight mode, that is, to conduct grid-based flight scanning of the entire planned area of ​​the power plant. For a power plant with a huge area, construction activities often only occur in a few local areas during a specific period. Adopting a full coverage approach means that a lot of UAV flight time, battery life and subsequent data processing resources are wasted in a vast area that has not changed, resulting in low overall efficiency and high cost.

[0004] (2) The inherent passivity and lag of monitoring: Current monitoring technologies are essentially "reactive" or "retrospective," that is, recording and analyzing completed construction activities. This model cannot provide forward-looking information, and project managers can only passively respond to schedule deviations that have already occurred, rather than anticipating potential construction bottlenecks, optimizing resource allocation, or proactively planning the next phase of work. The lack of predictive capabilities keeps project management in a constant state of "catching up."

[0005] (3) Insufficient fusion of multi-source data and lack of synergy: Although satellite remote sensing and UAV monitoring have been applied to the engineering field as independent technologies, a systematic approach to deeply integrate the two in an intelligent and collaborative manner to serve construction progress surveys is still immature.

[0006] Existing patented technologies are mostly concentrated on single-platform applications, such as using drones for photovoltaic module identification and defect detection, but they have failed to form an efficient collaborative workflow of "satellite macro-guidance and drone precise detailed inspection." The barriers between data sources have led to a "1+1<2" situation, failing to fully leverage the advantages of each.

[0007] This patent, published in China (CN119027069B) on February 25, 2025, discloses a method for identifying the construction progress of photovoltaic projects based on UAV imagery and neural networks. The patent primarily focuses on using UAV imagery and artificial intelligence algorithms to identify and extract components such as photovoltaic panels, modules, and piles during the construction of photovoltaic power plants. However, this patent cannot efficiently identify the construction progress of large-scale photovoltaic power plants, resulting in an inefficient investigation method.

[0008] This patent, published in China (CN118038225B) on March 25, 2025, discloses a photovoltaic (PV) panel extraction technology based on satellite imagery. It provides a method for PV panel identification using satellite imagery as a data source and combining a SAM (Search Engine Analysis) model with a SVM (Search Engine Visualization) classification model. This patent demonstrates the technical feasibility of using satellite imagery for PV panel identification. However, considering image accuracy, it cannot identify individual PV panel modules. Summary of the Invention

[0009] The purpose of this invention is to provide a method for investigating the construction progress of photovoltaic power plants using a combination of satellites and drones, in order to solve the technical problem of low efficiency in existing investigation methods that cover large areas of photovoltaic power plants with drones.

[0010] To address the aforementioned technical problems, this invention provides a method for investigating the construction progress of photovoltaic power plants using a satellite and drone collaboration, comprising: S1. Based on the set conditions of the photovoltaic power station, perform a remote sensing dataset search. If a remote sensing image of the photovoltaic power station that meets the set conditions is found, proceed to step S2. S2. The multi-band image data in the remote sensing image data of the photovoltaic power station are fused to generate image data; S3. Based on the image data from step S2, extract the photovoltaic panel target from the image and generate a closed photovoltaic panel installation area; S4. By overlaying the recognition results of the images of the photovoltaic panel installation areas of the two adjacent phases, the latest construction area is calculated. S5. The latest construction area calculated in step S4 is taken as the construction target area; high-precision drone image data collection is carried out in the construction target area using drones. S6. Based on the high-precision UAV image data collected in step S5, generate a high-precision orthophoto map, extract component-level targets including photovoltaic panels, supports, and piles from the high-precision orthophoto map, and count and identify the number and spatial location of component-level targets. S7. Filter duplicate component-level targets and update the number of component-level targets; S8. Based on the target number of completed construction components and the planned design component number, conduct statistical analysis to obtain construction progress.

[0011] First preferred option: If no photovoltaic power station remote sensing image that meets the set conditions is found in step S1, the prediction model is invoked to predict the possible construction target area for the next step based on the historical construction progress; and high-precision UAV image data of the construction target area is collected using UAVs. Then proceed with steps S6-S7.

[0012] Second preferred option: Predict the possible next construction target areas, specifically: Calculate the bounding rectangle of the construction area at time T-1, obtain the coordinate range of the construction area at time T-1, and calculate the geometric center of the area; Calculate the bounding rectangle of the construction area at time T, obtain the coordinate range of the construction area at time T, and calculate the geometric center of the area; Based on the obtained geometric centers of the two regions, the construction trend vector is calculated as the predicted direction for the next construction phase. The predicted construction speed is calculated based on the construction speed and the historical average speed. Based on the construction speed and interval, calculate the predicted installation volume within this time range; based on the boundary area of ​​the photovoltaic panels installed in the previous stage of the design, select the photovoltaic panels that may have been installed, calculate the circumscribed rectangle of the photovoltaic panel, and use it as the predicted construction target area.

[0013] Third preferred option: Step S3, in detail: S31. Based on the image data generated in step S2, use a deep learning AI target recognition algorithm to identify pixels in the image data that belong to the photovoltaic panel category; S32. Based on the above identification results, generate the polygonal boundary of the photovoltaic panel coverage area as a closed photovoltaic panel installation area.

[0014] Fourth priority option: In step S4, the latest construction area is calculated, specifically: The photovoltaic panel installation area extracted in step S3 for the current period is overlaid with the photovoltaic panel installation area extracted in step S3 for the previous period, and the newly added photovoltaic panel area is counted as the latest construction area.

[0015] Fifth priority option: In step S6, a deep learning AI target recognition algorithm is used to extract component-level targets, including photovoltaic panels, supports, and piles, from the high-precision orthophoto image.

[0016] Sixth priority option: Step S6, in detail: S61. Based on the high-precision UAV image data collected in step S5, use the YOLO V11 artificial intelligence model to identify target at the component level, including photovoltaic panels, modules, and piles, and extract the bounding box image coordinates of the target at the component level, as well as the coordinates of its center point. S62. Based on the geographic coordinate information in the high-precision UAV image data, calculate the geographic coordinates of the target bounding box and center point at the identification component level.

[0017] Seventh priority option: In step S7, duplicate component-level targets are filtered out: the newly extracted component-level targets in step S6 are spatially matched with the archived component-level targets in the database to filter out duplicate targets.

[0018] Eighth priority option: Step S7, specifically, uses spatial location filtering to remove duplicates from the target: First, the newly collected data is classified into three categories: photovoltaic panels, modules, and power stations. Then, the spatial distance between the categories and similar targets in the database is calculated. If the distance is less than the threshold D, the target is considered to be the same and is removed.

[0019] Ninth priority option: It also includes continuously optimizing the prediction model using newly calculated construction progress data.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects: This invention employs a collaborative approach combining remote sensing satellites and unmanned aerial vehicles (UAVs) to survey the construction progress of photovoltaic power plants. This method integrates the wide-area coverage capability of remote sensing satellites with the high-precision detail acquisition advantages of UAVs, constructing a "macro-micro" collaborative survey system. Satellite remote sensing can quickly acquire the overall construction outline of a photovoltaic power plant at a regional scale, promptly detect macro-level changes in the construction area, and determine the latest construction target area. UAVs, on the other hand, target the construction area identified in the satellite imagery, performing close-range, high-resolution data acquisition within that area, enabling accurate identification and quantity counting of construction targets such as photovoltaic panels, supports, and piles. This collaborative operation between satellites and UAVs effectively solves the problem of low efficiency in large-scale photovoltaic power plant operations with full UAV coverage.

[0021] Construction area prediction when satellite imagery is unavailable: By constructing a construction progress prediction model and combining historical construction speed, recent construction area, and recent construction speed, a scientific inference and prediction of construction area progress can be achieved during periods when satellite imagery is unavailable. Based on the prediction results, guidance for drone survey operations can continue, avoiding delays in construction supervision caused by gaps in satellite remote sensing data. This ensures the periodic and continuous conduct of progress surveys.

[0022] This invention utilizes a collaborative approach of remote sensing satellites and drones to survey the construction progress of photovoltaic power plants. First, satellite remote sensing imagery is used to identify photovoltaic panels, obtaining the current construction area of ​​the photovoltaic power plant on a macro scale. This information is then compared and analyzed with historical data to confirm new construction areas within the specified timeframe. Using this as the target area, drones are employed for small-scale, high-precision data collection and target identification. This further enables the accurate identification, positioning, and quantity counting of targets such as photovoltaic panels, supports, and piles. Finally, the data is compared with design drawings and quantities to achieve precise statistics on construction progress.

[0023] This invention provides a method for large-scale photovoltaic power plant construction progress survey using a combination of satellite and UAV collaboration. The method locates the construction area using satellite imagery and then utilizes UAVs for high-precision data acquisition and target identification. Ultimately, it achieves efficient, accurate, scalable, and forward-looking forecasting capabilities for large-scale photovoltaic power plant construction progress surveys.

[0024] This invention provides a method for large-scale photovoltaic power plant construction progress survey using a combination of satellite and UAV collaboration. The method employs a two-stage survey strategy: "coarse-to-fine, prediction-first." First, it utilizes remote sensing satellite imagery with wide coverage and short revisit cycles for change monitoring, quickly and cost-effectively identifying active construction areas throughout the power plant. Then, high-resolution, highly mobile UAVs are deployed only to the identified active construction areas for precise data collection. Deep learning algorithms automatically extract key construction targets such as photovoltaic panels, supports, and piles from the UAV imagery, obtaining their centimeter-level geographic coordinates. This "current status" data is compared with the project's geospatial database to accurately quantify the current construction progress.

[0025] This invention addresses the technical problems of existing survey methods, such as low efficiency, high cost, inherent passivity, and lack of data synergy, by shifting from conventional passive, delayed monitoring to proactive predictive surveys.

[0026] One of the core innovations of this invention lies in the introduction of a construction prediction model. This model analyzes construction progress data accumulated from previous surveys to uncover the spatial and temporal evolution patterns of construction activities, thereby predicting the "hotspot" areas most likely to be under construction in the next cycle. This prediction not only provides project managers with forward-looking decision support but also guides preliminary surveys during the "window period" before new satellite imagery data is available.

[0027] This invention constructs a closed-loop feedback and self-optimization system. When new satellite imagery data arrives, the system immediately compares and verifies it against the model's predicted areas. This process not only identifies and corrects potential omissions in the prediction model, ensuring the completeness of the survey (by triggering supplementary surveys of missed areas), but more importantly, it uses prediction biases (including missed and false alarm areas) as valuable training data, feeding them back to the prediction model to continuously improve its accuracy. This adaptive, self-learning mechanism enables the entire survey system to continuously evolve throughout the project cycle, becoming increasingly intelligent and accurate. Attached Figure Description

[0028] Figure 1 This is a flowchart of the method of the present invention.

[0029] Figure 2 This is a schematic diagram illustrating the process of predicting the construction area according to the present invention.

[0030] Figure 3 This is a flowchart illustrating the deduplication of the target space in this invention.

[0031] Figure 4 This is a region map of the photovoltaic panel range extracted from remote sensing images according to the present invention, which is the area enclosed by the solid green line in the figure.

[0032] Figure 5 The area map for identifying the construction area of ​​this invention is the area enclosed by the yellow solid line in the figure.

[0033] Figure 6 This is a diagram showing the target recognition result of the UAV imagery according to the present invention. Detailed Implementation

[0034] The following non-limiting embodiments are intended to enable those skilled in the art to gain a more comprehensive understanding of the present invention, but do not limit the invention in any way. The following content is merely an exemplary description of the scope of protection claimed by the present invention, and those skilled in the art can make various changes and modifications to the present invention based on the disclosed content, and such changes should also fall within the scope of protection claimed by the present invention.

[0035] This invention proposes a satellite-and-UAV collaborative method for investigating the construction progress of photovoltaic power plants, employing a "coarse-to-fine, prediction-first" strategy. First, leveraging the wide coverage and short cycle of remote sensing satellite imagery, active construction areas of the power plants are quickly identified. Then, UAVs are dispatched to these areas to collect high-precision data. Deep learning algorithms are used to extract the geographic coordinates of key targets, which are then compared with planned construction volumes to achieve quantitative analysis of construction progress. The core innovation is a construction prediction model that analyzes historical data to predict construction hotspots, providing forward-looking decision support and filling gaps in satellite data updates. Simultaneously, a closed-loop feedback system is constructed. New satellite imagery is compared with the predicted areas for verification, correcting prediction errors. Deviation data is fed back to the model to optimize prediction accuracy, enabling continuous system evolution and improving intelligence and accuracy. The main implementation process is as follows: Figure 1 .

[0036] like Figure 1 As shown, this invention provides a method for investigating the construction progress of photovoltaic power plants using a combination of satellite and drone, including: S1. Remote Sensing Data Retrieval: Based on the set conditions of the photovoltaic power station, a remote sensing dataset is retrieved. If a remote sensing image of the photovoltaic power station that meets the set conditions is found, proceed to step S2. The conditions set include the location and spatial range of the photovoltaic power station, the survey time, cloud cover, and image spatial resolution. S2. Remote sensing image data acquisition: If the remote sensing image data that meets the requirements is queried and obtained in step S1, then the multi-band image data in the remote sensing image data of the photovoltaic power station is fused to generate image data. S3. Identification and extraction of photovoltaic panel installation areas: Based on the image data from step S2, a deep learning AI target recognition algorithm is used to extract photovoltaic panel targets from the image and generate a closed photovoltaic panel installation area; specifically: S31. Based on the image data generated in step S2, use a deep learning AI target recognition algorithm to identify pixels in the image data that belong to the photovoltaic panel category; S32. Based on the above identification results, generate the polygonal boundary of the photovoltaic panel coverage area as a closed photovoltaic panel installation area.

[0037] S4. Analysis of Changes in the Construction Area: By overlaying and analyzing the recognition results of images from two adjacent photovoltaic panel installation areas, the latest construction area can be calculated. Latest construction area calculations, details: The photovoltaic panel installation area extracted in step S3 for the current period is overlaid with the photovoltaic panel installation area extracted in the previous period for analysis, and the newly added photovoltaic panel area is counted as the latest construction area.

[0038] S5, Detailed drone survey: The latest construction area calculated in step S4 is taken as the construction target area; high-precision drone image data acquisition with a spatial resolution better than 10cm is carried out in the construction target area using drones. S6. Target recognition of photovoltaic power plants using drone imagery: Based on the high-precision UAV image data collected in step S5, a high-precision orthophoto map is generated. Using a deep learning AI target recognition algorithm, component-level targets such as photovoltaic panels, supports, and piles are extracted from the high-precision orthophoto map. The number and spatial location of the identified component-level targets are also counted. Step S6, in detail: S61. Based on the high-precision UAV image data collected in step S5, use the YOLO V11 artificial intelligence model to identify target at the component level, including photovoltaic panels, modules, and piles, and extract the bounding box image coordinates of the target at the component level, as well as the coordinates of its center point. S62. Based on the geographic coordinate information in the high-precision UAV image data, calculate the geographic coordinates of the target bounding box and center point at the identification component level.

[0039] S7. Incremental update of construction data: Since it is inevitable that the UAV will collect data from the surveyed area during the data collection process, in order to achieve accurate data statistics, the newly extracted component-level targets in step S6 need to be matched with the archived component-level targets in the database to filter duplicate component-level targets and update the number of component-level targets. Specifically, spatial location filtering is used for target deduplication: First, the newly collected data is classified into three categories: photovoltaic panels, modules, and power stations. Then, the spatial distance between the categories and similar targets in the database is calculated. If the distance is less than the threshold D, the target is considered to be the same and is removed.

[0040] S8. Construction Progress Statistics: After completing the data update in step S7, statistical analysis is performed on the target number of completed construction components and the planned design component number to obtain the construction progress.

[0041] S9. Optimize the prediction model: Because there may be situations where no remote sensing imagery is available during the survey period, such as out-of-coverage during the revisit period or cloud cover, the prediction model is continuously optimized using the construction progress data newly calculated in step S8 to improve its prediction accuracy.

[0042] If no photovoltaic power station remote sensing image that meets the set conditions is found in step S1, the prediction model is invoked to predict the possible construction target area for the next step based on the historical construction progress; and high-precision UAV image data of the construction target area is collected using UAVs. Then proceed with steps S6-S7.

[0043] Predict the possible next construction target areas, specifically: Calculate the circumscribed rectangle of the photovoltaic power station construction area at time T-1, obtain the coordinate range of the photovoltaic power station construction area at time T-1, and calculate the geometric center of the area; Calculate the circumscribed rectangle of the photovoltaic power station construction area at time T, obtain the coordinate range of the photovoltaic power station construction area at time T, and calculate the geometric center of the area; Based on the obtained geometric centers of the two regions, the construction trend vector is calculated as the predicted direction for the next construction phase. The predicted construction speed is calculated based on the construction speed and the historical average speed. Based on the construction speed and interval time, calculate the predicted installation volume within the interval time range; based on the boundary area of ​​the photovoltaic panels installed in the previous stage of the design, select the photovoltaic panels that may have been installed, calculate the circumscribed rectangle of the photovoltaic panel, and use it as the predicted construction target area.

[0044] Example This paper provides a specific embodiment of a method for investigating the construction progress of photovoltaic power plants using a combination of satellite and drones. For example... Figure 3 As shown, the specific steps include the following: S1. Remote sensing data retrieval: S11. First, determine the spatial range, time range, cloud cover limit (e.g., <5%), and satellite type (e.g., Sentinel-2, Jilin-1 satellite, etc.) of the photovoltaic power station to be investigated.

[0045] Since the primary purpose of satellite imagery is to determine the construction area, rather than to accurately identify targets such as photovoltaic panels, supports, and piles, Sentinel-2 satellites with a 10-meter resolution can be used as the data source. The data can be downloaded free of charge from the European Space Agency website (https: / / browser.dataspace.copernicus.eu / ). This invention uses the Zhongwei photovoltaic power station as an example for data collection. The search criteria are set as follows: Data selection: SENTINEL-2, MSI sensor, L2A level data. Visible light B4, B3, and B2 band combination.

[0046] Spatial range: [36.95, 105.006] - [37.95, 106.25]; Time range: May 18, 2025 - May 28, 2025; Cloud cover requirement: <10%.

[0047] S2. Remote sensing image data acquisition: S21. Retrieve data that meets the requirements from the remote sensing image data provider according to the set conditions; S22. If the area to be investigated is distributed across several images, use remote sensing data processing software to mosaic and crop the image data. S23. After completing the above processing, perform coordinate transformation to convert the image to the target projection coordinate system and generate image data.

[0048] S3. Identification and extraction of photovoltaic panel installation areas: S31. Based on the image generated in step S2, use random forest and its learning and recognition algorithm to identify pixels in the image that belong to the photovoltaic panel category; S32. Based on the above identification results, generate the polygonal boundary of the photovoltaic panel coverage area.

[0049] S4. Analysis of Changes in the Construction Area: The photovoltaic panel installation area (denoted as AT) extracted in step S3 for the current period is overlaid with the photovoltaic panel installation area (denoted as AT-1) extracted in step S2 of the previous period, and the newly added photovoltaic panel area (denoted as ΔA) is counted as the latest construction target area.

[0050] Based on Google Earth Engine (GEE), the random forest algorithm was used to identify and extract photovoltaic panels. First, two images of the Sentinel-2 Zhongwei Power Station taken on May 18 and May 28 were loaded. Then, training samples of photovoltaic panels and non-photovoltaic panels were prepared (labeled as 1 and 0, respectively), merged, and divided into training and test sets in a 7:3 ratio. The built-in random forest classifier of GEE (100 trees, 0.7 bag score) was used for training to perform classification on the feature set. Small patches smaller than 100 square meters were removed using connected component analysis to optimize the results; overall accuracy, Kappa coefficient, and confusion matrix were calculated to assess classification accuracy; finally, true-color imagery, initial classification results, and final photovoltaic panel extraction results were visualized and exported locally. These results were then processed using QGIS (a free and open-source cross-platform desktop geographic information system software) to generate the photovoltaic panel coverage area. Finally, the extraction results from two date images were overlaid and analyzed to extract the changed areas as construction areas. Figure 4 , 5 As shown.

[0051] S5, Detailed drone survey: S51. Determine the target area: If available satellite imagery is available, the target construction area is ΔA calculated in step S4. If available satellite imagery is unavailable, the construction target area will be predicted based on the following prediction model, such as... Figure 2 As shown: Calculate the circumscribed rectangle of the construction area at time T-1, and obtain the coordinate range of the construction area [Xminp,Yminp], [Xmaxp,Ymaxp], where [Xminp,Yminp] is the coordinate of the lower left corner of the circumscribed rectangle, and [Xmaxp,Ymaxp] is the coordinate of its upper right corner; and calculate the geometric center of the region P[(Xminp+Xmaxp) / 2, (Yminp+Ymaxp) / 2]. It is a change over time (e.g., 1 day).

[0052] The same method is used to calculate the construction range [Xminq, Yminq] and [Xmaxq, Ymaxq] at time T, where [Xminq, Yminq] are the coordinates of the lower left corner of the circumscribed rectangle, and [Xmaxq, Ymaxq] are the coordinates of its upper right corner and its geometric center Q [(Xminq+Xmaxq) / 2, (Yminq+Ymaxq) / 2]. The construction trend vector pq is calculated as the predicted direction for the next construction step.

[0053] Simultaneously, the construction speed per unit time is calculated. Since the layout of photovoltaic panels in a photovoltaic power station is not necessarily regular, especially in mountainous areas, the speed is measured in the number of photovoltaic panels installed per unit time. Considering the impact of the construction area on the construction speed, the most recent construction speed and the historical average speed are calculated, and then the maximum speed between the two is taken as the predicted construction speed.

[0054] The most recent construction speed Vr is calculated as follows: the number of installations completed at time Tt is CT. At times t and T, the number of installations completed is CT, and the construction speed is VR = (CT / T). CT t ) / t.

[0055] The historical average speed is VA = C / T, where C is the total amount of installation completed and T is the total construction time. The final predicted construction speed is V = max(VR,VA).

[0056] Based on the construction speed V and the interval time t, calculate the predicted installation volume F=Vt within this time range. Then, based on the design drawings and the boundary area completed in the previous stage, select the photovoltaic panels that may be installed from the drawings and calculate the circumscribed rectangle of the photovoltaic panel as the construction target area ΔA.

[0057] S52. UAV data acquisition: After determining the construction area ΔA, the UAV performs a flight survey mission to collect high-precision orthophoto data of the target area.

[0058] S6, Target Recognition of Photovoltaic Power Stations from Drone Images S61, Target Recognition: Based on the high-precision UAV imagery collected in S5, the YOLO V11 artificial intelligence model is used to identify targets such as photovoltaic panels, modules, and piles, and to extract the bounding box image coordinates of the targets, as well as the coordinates of their center points.

[0059] S62. Spatial positioning: Calculate and identify the geographic coordinates of the target bounding box and center point based on the geographic coordinate information in the image.

[0060] After confirming the construction area, a DJI M30T drone was used to collect data on the area. Then, Pix4d software was used to generate orthophotos. To improve recognition efficiency, the images were formatted at 512 pixels. The data is segmented into 512 pixels. YOLO v11, learned through transfer learning from a photovoltaic power station dataset, is used as the recognition model to identify photovoltaic panels, supports, and piles. Testing showed a recognition efficiency better than 30 FPS, an accuracy >98%, and a target spatial positioning accuracy <10cm. This enables accurate identification and positioning of photovoltaic panels, supports, and piles.

[0061] S7, Incremental Data Update for Construction Data S71. Data Deduplication: To address the issue of duplicate target identification caused by spatial overlap between the UAV image capture area and the pre-statistical area, data deduplication is performed. Since the positioning accuracy of UAV orthophotos can reach the centimeter level, while the size of targets such as photovoltaic panels, modules, and piles is on the meter level, a spatial location filtering method is used for target deduplication. The specific method is as follows... Figure 3 As shown: First, the newly collected data is classified into three categories: photovoltaic panels, modules, and station piles. Then, the spatial distance between the categories and similar targets in the database is calculated. If the distance is less than the threshold D, the target is considered to be the same target and is removed from the set.

[0062] S72. Data Import: After deduplication, the newly collected construction target data will be imported into the photovoltaic synchronization database to update the data.

[0063] After identification, spatial filtering is performed using 20cm as a threshold and the targets stored in the database to remove duplicate target objects. Incremental data entry and database updates are then performed. Figure 6 As shown.

[0064] S8, Construction Progress Statistics After the S7 data update is completed, statistical analysis is performed based on the number of completed slabs, columns, piles, and supports in the database and the number of components in the planned design to obtain the construction progress.

[0065] S9. Optimize the prediction model Because there may be situations where no remote sensing imagery is available during the survey period, such as out-of-coverage during the revisit period or cloud cover, the prediction model is continuously optimized using newly calculated construction progress data to improve its prediction accuracy.

[0066] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for investigating the construction progress of photovoltaic power plants using a combination of satellite and drones, characterized in that... include: S1. Based on the set conditions of the photovoltaic power station, perform a remote sensing dataset search. If a remote sensing image of the photovoltaic power station that meets the set conditions is found, proceed to step S2. S2. The multi-band image data in the remote sensing image data of the photovoltaic power station are fused to generate image data; S3. Based on the image data from step S2, extract the photovoltaic panel target from the image and generate a closed photovoltaic panel installation area; S4. By overlaying and analyzing the recognition results of the images of the photovoltaic panel installation areas of the two adjacent phases, the latest construction area can be calculated. S5. Take the latest construction area calculated in step S4 as the construction target area; or construct a prediction model and use the prediction model to predict the construction target area; use drones to collect high-precision drone image data of the construction target area. S6. Based on the high-precision UAV image data collected in step S5, generate a high-precision orthophoto map, extract component-level targets including photovoltaic panels, supports, and piles from the high-precision orthophoto map, and count and identify the number and spatial location of component-level targets. S7. Filter duplicate component-level targets and update the number of component-level targets; S8. Based on the target number of completed construction components and the planned design component number, conduct statistical analysis to obtain construction progress.

2. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, If no photovoltaic power station remote sensing image that meets the set conditions is found in step S1, the prediction model is invoked to predict the possible construction target area for the next step based on the historical construction progress; and high-precision UAV image data of the construction target area is collected using UAVs. Then proceed with steps S6-S7.

3. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 2, characterized in that, Predict the possible next construction target areas, specifically: Calculate the bounding rectangle of the construction area at time T-1, obtain the coordinate range of the construction area at time T-1, and calculate the geometric center of the area; Calculate the bounding rectangle of the construction area at time T, obtain the coordinate range of the construction area at time T, and calculate the geometric center of the area; Based on the obtained geometric centers of the two regions, the construction trend vector is calculated as the predicted direction for the next construction phase. The predicted construction speed is calculated based on the construction speed and the historical average speed. Based on the construction speed and interval, calculate the predicted installation volume within this time range; based on the boundary area of ​​the photovoltaic panels installed in the previous stage of the design, select the photovoltaic panels that may have been installed, calculate the circumscribed rectangle of the photovoltaic panel, and use it as the predicted construction target area.

4. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, Step S3, in detail: S31. Based on the image data generated in step S2, use a deep learning AI target recognition algorithm to identify pixels in the image data that belong to the photovoltaic panel category; S32. Based on the above identification results, generate the polygonal boundary of the photovoltaic panel coverage area as a closed photovoltaic panel installation area.

5. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, In step S4, the latest construction area is calculated, specifically: The photovoltaic panel installation area extracted in step S3 for the current period is overlaid with the photovoltaic panel installation area extracted in step S3 for the previous period, and the newly added photovoltaic panel area is counted as the latest construction area.

6. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, In step S5, the prediction model is constructed, specifically: Calculate the bounding rectangle of the construction area at time T-1; calculate the bounding rectangle of the construction area at time T; calculate the construction trend vector at both times as the predicted direction of the next construction step; Calculate the construction speed per unit time to obtain the predicted construction speed; Based on the predicted construction speed and interval, calculate the predicted number of photovoltaic panels to be installed within the time range. Based on the boundary area completed in the previous stage, select the photovoltaic panels that have been installed and calculate the circumscribed rectangle of the photovoltaic panel as the construction target area.

7. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, In step S6, a deep learning AI target recognition algorithm is used to extract component-level targets, including photovoltaic panels, supports, and piles, from the high-precision orthophoto image.

8. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 7, characterized in that, Step S6, in detail: S61. Based on the high-precision UAV image data collected in step S5, use the YOLO V11 artificial intelligence model to identify target at the component level, including photovoltaic panels, modules, and piles, and extract the bounding box image coordinates of the target at the component level, as well as the coordinates of its center point. S62. Based on the geographic coordinate information in the high-precision UAV image data, calculate the geographic coordinates of the target bounding box and center point at the identification component level.

9. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, In step S7, duplicate component-level targets are filtered out: the newly extracted component-level targets in step S6 are spatially matched with the archived component-level targets in the database to filter out duplicate targets.

10. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 8, characterized in that, Step S7, specifically, uses spatial location filtering to remove duplicates from the target: First, the newly collected data is classified into three categories: photovoltaic panels, modules, and power stations. Then, the spatial distance between the categories and similar targets in the database is calculated. If the distance is less than the threshold D, the target is considered to be the same and is removed.

11. The method for investigating the construction progress of a photovoltaic power station using satellite and UAV collaboration as described in claim 1, characterized in that, It also includes continuously optimizing the prediction model using newly calculated construction progress data.

Images