Photovoltaic power station construction progress monitoring method

Abstract

The invention relates to a photovoltaic power station construction progress monitoring method based on unmanned aerial vehicle aerial video, and the monitoring method comprises the following steps: S1, building stage division and data set establishment; s2, improving a YOLOv8 target detection model; s3, model training and key component detection; s4, counting key components based on the video; s5, construction stage determination and progress calculation; the construction process of the photovoltaic power station is analyzed, the target detection algorithm is improved, the key components in the construction process of the photovoltaic power station are detected, the construction progress information of the photovoltaic power station of the target land parcel is obtained according to the type and the number of the detected key components, visualization, refinement, standardization and intelligentization of construction management of the photovoltaic power station are achieved, and the construction efficiency of the photovoltaic power station is improved. The construction quality and comprehensive benefits of a photovoltaic power station project are effectively improved, the problem that the construction progress statistical efficiency of an existing photovoltaic power station is low is solved, and the method has great significance in promoting digital and intelligent management of the photovoltaic power station project.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station construction, and in particular to a method for monitoring the construction progress of a photovoltaic power station. Background Art

[0002] In the context of vigorously developing a new power system with new energy as the main energy source, photovoltaic power generation, as an important part of clean energy, has received strong support and encouragement from the government.

[0003] As the scale of photovoltaic power generation base stations continues to expand and the number of construction projects continues to increase, construction progress control is becoming increasingly complex and difficult. These problems, including the large number of construction parties, complex and changing environments, and large amounts of data, lead to inefficient construction progress management. Traditional construction progress monitoring requires the construction party to manually report the daily construction progress to the system, which is a time-consuming, costly, and labor-intensive task with a low degree of automation.

[0004] Therefore, it is particularly important to realize automated progress monitoring and improve the efficiency of progress management in complex environments. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a method for monitoring the construction progress of a photovoltaic power station to solve the problem of low statistical efficiency of the construction progress of a photovoltaic power station in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for monitoring the construction progress of a photovoltaic power station based on drone aerial video. The method comprises the following steps:

[0007] S1, construction phase division and dataset establishment;

[0008] The construction phase division specifically divides the construction of photovoltaic power stations into four stages: concrete pipe pile construction, steel spiral pile construction, photovoltaic support construction, and photovoltaic panel laying. The key components of each stage are specified as concrete pipe piles, steel spiral piles, photovoltaic support, and photovoltaic panels.

[0009] S2, YOLOv8 target detection model improvement;

[0010] S3, model training and key component detection;

[0011] S4, video-based key component counting;

[0012] S5. Determination of construction phase and calculation of progress.

[0013] By adopting this technical solution, the construction process of photovoltaic power stations is analyzed and the target detection algorithm is improved. The key components in the construction process of photovoltaic power stations are detected. According to the type and number of key components detected, the construction progress information of the photovoltaic power station in the target plot is obtained, and the construction management of photovoltaic power stations is visualized, refined, standardized and intelligent. The construction quality and comprehensive benefits of photovoltaic power station projects are effectively improved, and the problem of low statistical efficiency of the existing photovoltaic power station construction progress is solved. It is of great significance to promote the digitalization and intelligent management of photovoltaic power station projects.

[0014] Furthermore, the dataset in step S1 is established by collecting aerial images corresponding to the four stages through open source images and real-life photography on the Internet. Labelme is used as the image annotation tool software, and the Polygen tool is used to perform instance annotation on the key components in the collected images to determine the outline and position of the key components, thereby forming an image dataset for identifying the construction progress of photovoltaic power stations.

[0015] By adopting this technical solution, the construction stage is associated with the key components, providing a basis for determining which construction stage the key component type is in based on the subsequent acquisition through photography.

[0016] Furthermore, step S2 includes improving the YOLOv8 model so that it can achieve accurate identification of key components in each construction stage in the case of target and background confusion and small targets.

[0017] By adopting this technical solution, the accuracy of obtaining the types and quantities of subsequent key components can be increased.

[0018] Furthermore, the YOLOv8 model improvement includes optimizing and streamlining the YOLOv8 model using model distillation technology.

[0019] By adopting this technical solution, knowledge transfer from a small model to a large model is achieved, meeting the strict computer resource constraints of embedded devices. This enables the YOLOv8 model to perform image analysis and processing directly on the drone side, increasing its applicability to meet the challenges of unstable network connections in remote areas such as deserts.

[0020] Furthermore, model distillation technology transfers the knowledge of a complex teacher model to a student model with a simpler structure and more efficient operation, including learning feature representations and imitating responsible decision boundaries.

[0021] By adopting this technical solution, the performance of the student model on small object detection tasks is improved.

[0022] Furthermore, step S3 includes using the photovoltaic power station construction progress recognition image dataset to train the improved YOLOv8 model, and applying the trained YOLOv8 model to perform target recognition on the photovoltaic power station aerial images.

[0023] By adopting this technical solution, the trained YOLOv8 model has higher recognition accuracy to increase the recognition effect of key components.

[0024] Furthermore, step S4 includes applying the trained YOLOv8 model to analyze the aerial video of the construction site and counting the number of key components appearing in the video.

[0025] Furthermore, the Deepsort target tracking algorithm is introduced to receive the real-time aerial video stream from the drone, and then the trained YOLOv8 model is used to identify the key components and their quantity in the picture.

[0026] Furthermore, the virtual coil method is used to count key components. By setting virtual detection coils in the aerial video, the number of components passing through these coils is automatically calculated. When the components tracked by Deepsort pass through the set virtual coils, the system automatically counts and updates the construction progress data in real time.

[0027] By adopting this technical solution, duplicate counting is avoided based on Deepsort tracking results, and the statistical accuracy of the number of key components is increased.

[0028] Furthermore, step S5 includes determining the construction stage of the video shooting site based on the analysis result of the construction site aerial video, and calculating the construction progress percentage of the stage.

[0029] By adopting this technical solution, the current construction progress of the photovoltaic power station is obtained.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. Analyze the construction process of photovoltaic power stations and improve the target detection algorithm. Detect key components in the construction process of photovoltaic power stations. According to the type and number of key components detected, obtain the construction progress information of photovoltaic power stations in the target plot. Realize the visualization, refinement, standardization and intelligence of photovoltaic power station construction management, effectively improve the construction quality and comprehensive benefits of photovoltaic power station projects, solve the problem of low efficiency of existing photovoltaic power station construction progress statistics, and have important significance for promoting the digitalization and intelligent management of photovoltaic power station projects.

[0032] 2. Define model distillation technology to achieve knowledge transfer from small models to large models, meet the strict computing resource constraints of embedded devices, and enable the model to perform image analysis and processing directly on the drone end to address the challenges of unstable network connections in remote areas such as deserts. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of the process of monitoring the construction progress of a photovoltaic power station in the present invention;

[0034] Figure 2 Schematic diagram of the model distillation technology principle in the photovoltaic power station construction progress monitoring method of the present invention;

[0035] Figure 3 A schematic diagram showing the counting effect of key components of the photovoltaic power station construction progress monitoring method of the present invention;

[0036] Figure 4 Schematic diagram of the key component counting process of the photovoltaic power station construction progress monitoring method of the present invention;

[0037] Figure 5 This is a construction progress determination and calculation logic diagram for the photovoltaic power station construction progress monitoring method in the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0039] Tool Description: Labelme is an image annotation software; Polygen is a polygon annotation tool; YOLOv8 is a model software; Deepsort is a multi-target tracking algorithm;

[0040] Please see the attached Figure 1 The present invention provides a method for monitoring the construction progress of a photovoltaic power station based on drone aerial video. The monitoring method comprises the following steps:

[0041] S1, construction phase division and dataset establishment;

[0042] Among them, the construction stage division is to divide the construction of photovoltaic power stations into four stages: concrete pipe pile construction, steel spiral pile construction, photovoltaic bracket construction and photovoltaic panel laying. The key components of each stage are specified as concrete pipe piles, steel spiral piles, photovoltaic brackets and photovoltaic panels respectively; the dataset is established by collecting aerial images corresponding to the four stages through open source images and real-life shooting on the Internet, applying Labelme as the image annotation tool software, and using the Polygen tool to instance-annotate the key components in the collected images, determine the outline and position of the key components, and form an image dataset for identifying the construction progress of photovoltaic power stations.

[0043] S2, YOLOv8 target detection model improvement;

[0044] This includes improvements to the YOLOv8 model, enabling it to accurately identify key components at each construction stage even in situations where the target is confused with the background and small objects are present;

[0045] For further information, please see the attached Figure 2 Improvements to the YOLOv8 model include optimizing and streamlining the model using model distillation technology; enabling knowledge transfer from a small model to a large model, meeting the strict computer resource constraints of embedded devices, and enabling the YOLOv8 model to perform image analysis and processing directly on the drone side, increasing its applicability to address the challenges of unstable network connections in remote areas such as deserts;

[0046] Furthermore, model distillation technology transfers the knowledge of a complex teacher model to a student model with a simpler structure and more efficient operation, including learning feature representations and imitating responsible decision boundaries, thereby improving the performance of the student model on small target detection tasks.

[0047] S3, model training and key component detection;

[0048] Please see the attached Figure 4 , including using the photovoltaic power station construction progress recognition image dataset to train the improved YOLOv8 model, and applying the trained YOLOv8 model to perform target recognition on photovoltaic power station aerial images.

[0049] S4, video-based key component counting;

[0050] Continue to refer to the attached Figure 4 , including applying the trained YOLOv8 model to analyze aerial videos of the construction site and counting the number of key components appearing in the video;

[0051] Furthermore, the Deepsort target tracking algorithm is introduced to receive the real-time aerial video stream from the drone, and then the trained YOLOv8 model is used to identify the key components and their quantity in the picture.

[0052] Furthermore, the virtual coil method is used to count key components. By setting virtual detection coils in the aerial video, the number of components passing through these coils is automatically calculated. When the components tracked by Deepsort pass through the set virtual coils, the system automatically counts and updates the construction progress data in real time.

[0053] S5. Construction phase determination and progress calculation;

[0054] Please see the attached Figure 5, including determining the construction stage of the video shooting site based on the analysis results of the aerial video of the construction site, and calculating the construction progress percentage of that stage; and obtaining the current construction progress of the photovoltaic power station.

[0055] In step S2 of the above embodiment, in order to address the challenge of unstable network connection in remote areas such as deserts, an efficient model distillation technology is used to optimize and streamline the YOLOv8 model so that the model can be configured on embedded devices such as drones for offline image analysis and processing. Figure 4 As shown, the specific implementation steps include:

[0056] 1. Define a complex teacher model (e.g., YOLOv8x) with high accuracy but high computational cost. Due to its high computational requirements, this model is not suitable for direct application on embedded devices with limited computing power.

[0057] 2. Design a student model with a simplified structure and reduced computational requirements (e.g., YOLOv8s). The goal is to achieve or approach the performance standards of the teacher model while minimizing computational complexity.

[0058] 3. Using model distillation techniques to transfer knowledge from the teacher model to the student model, including feature representation learning and imitation of complex decision boundaries, to improve the student model's performance on fine-grained tasks such as small object detection;

[0059] 4. During the distillation process, the network structure and parameters of the student model are optimized to ensure that it can run effectively on embedded devices with limited computing power and maintain high-precision detection capabilities for small objects;

[0060] Through the above steps, not only the computational and storage burden of the model is significantly reduced, but also the model can run efficiently in a resource-limited environment, providing an effective solution for implementing deep learning models on embedded devices.

[0061] In step S4 of the above embodiment, the improved YOLOv8 model and DeepSort target tracking algorithm, as well as the virtual coil method, are combined to achieve automatic identification, tracking and counting of key components on the construction site. Figure 4 As shown, the specific implementation steps include:

[0062] 1. Use the trained YOLOv8 model to process construction site videos captured by drones. The model can accurately identify key construction components in the video, such as photovoltaic panels and brackets.

[0063] 2. Deepsort algorithm is integrated to extract component features through a deep learning model and use a Kalman filter to predict and update the component's motion trajectory, enabling the algorithm to continuously and effectively track the build;

[0064] 3. Set a virtual coil in the video frame as the detection area for component counting;

[0065] 4. When a component tracked by the Deepsort algorithm crosses the virtual coil, the system automatically triggers a counting mechanism. Each time a component is crossed, the counter is incremented, thus achieving real-time statistics on the number of key components.

[0066] 5. The system will update the data obtained by virtual coil counting to the construction progress database in real time, providing immediate progress feedback for project management.

[0067] In step S5 of the above embodiment, an algorithm logic is designed to accurately determine the specific stage of the construction site after obtaining the aerial video analysis results, and calculate the construction progress percentage of that stage, such as Figure 5 As shown, it specifically includes three steps:

[0068] 1. Obtain video analysis results, including the categories and number of key components contained in a video. In actual construction, the construction sequence for the same work area is to first lay the pile foundation (concrete pipe piles or steel screw piles), then install the photovoltaic brackets on the pile foundations, and finally lay the photovoltaic panels on the photovoltaic brackets. The second process will not begin until the first process is completed. Therefore, the same video will only contain two types of key components, namely the key component category of the current construction and the key component category corresponding to the previous process.

[0069] 2. Determine the construction stage of the site; based on the video analysis results in the first step, the detected component categories are divided into three situations: situation one, there are only pile foundation components (concrete pipe piles or steel spiral piles) on the site; situation two, the site contains pile foundation components and photovoltaic brackets; situation three, the site contains photovoltaic brackets and photovoltaic panels; when the detection result is situation one, it is considered that the work area is in the pile foundation laying stage; when the detection result is situation two, it is considered that the work area is in the photovoltaic bracket installation stage; when the detection result is situation three, it is considered that the work area is in the photovoltaic panel laying stage.

[0070] 3. Determine the construction progress percentage for the current stage. Based on the current construction stage determined in step 2, extract the number of corresponding components detected in the video. Divide this number by the planned number of photovoltaic panels to obtain the construction progress percentage for the current stage of the site.

[0071] The present invention has been described in detail above with reference to the embodiments of the accompanying drawings. A person skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention. The scope of protection of the present invention shall be determined by the scope defined in the appended claims.

Claims

1. A method for monitoring the construction progress of a photovoltaic power station based on drone aerial video, characterized in that: The monitoring method comprises the following steps: S1, construction phase division and dataset establishment; The construction phase division specifically divides the construction of photovoltaic power stations into four stages: concrete pipe pile construction, steel spiral pile construction, photovoltaic support construction, and photovoltaic panel laying. The key components of each stage are specified as concrete pipe piles, steel spiral piles, photovoltaic support, and photovoltaic panels. S2, YOLOv8 target detection model improvement; S3, model training and key component detection; S4, video-based key component counting; S5. Determination of construction phase and calculation of progress.

2. The photovoltaic power station construction progress monitoring method according to claim 1, characterized in that: The dataset in step S1 is established by collecting aerial images corresponding to the four stages through open source images and real-life photography on the Internet. Labelme is used as the image annotation tool software, and the Polygen tool is used to annotate the key components in the collected images, determine the outlines and positions of the key components, and form an image dataset for identifying the construction progress of photovoltaic power stations.

3. The method for monitoring the construction progress of a photovoltaic power station according to claim 2, characterized in that: Step S2 includes improving the YOLOv8 model so that it can accurately identify key components in each construction stage in the case of target-background confusion and small targets.

4. The method for monitoring the construction progress of a photovoltaic power station according to claim 3, characterized in that: YOLOv8 model improvements include optimizing and streamlining the YOLOv8 model using model distillation technology.

5. The method for monitoring the construction progress of a photovoltaic power station according to claim 4, characterized in that: Model distillation technology is to transfer the knowledge of a complex teacher model to a student model with a simpler structure and more efficient operation, including learning feature representation and imitating the responsible decision boundary.

6. The method for monitoring the construction progress of a photovoltaic power station according to claim 3, characterized in that: Step S3 includes using the photovoltaic power station construction progress recognition image dataset to train the improved YOLOv8 model, and applying the trained YOLOv8 model to perform target recognition on the photovoltaic power station aerial image.

7. The method for monitoring the construction progress of a photovoltaic power station according to claim 6, characterized in that: Step S4 includes applying the trained YOLOv8 model to analyze the aerial video of the construction site and counting the number of key components appearing in the video.

8. The method for monitoring the construction progress of a photovoltaic power station according to claim 7, characterized in that: The Deepsort target tracking algorithm is introduced to receive real-time aerial video streams from drones, and then the trained YOLOv8 model is used to identify key components and their quantity in the picture.

9. The method for monitoring the construction progress of a photovoltaic power station according to claim 8, characterized in that: The virtual coil method is used to count key components. By setting virtual detection coils in the aerial video, the number of components passing through these coils is automatically calculated. When the components tracked by Deepsort pass through the set virtual coils, the system automatically counts them and updates the construction progress data in real time.

10. The photovoltaic power station construction progress monitoring method according to claim 9, characterized in that: Step S5 includes determining the construction stage of the video shooting site based on the analysis results of the construction site aerial video, and calculating the construction progress percentage of the stage.

Images