Engineering construction progress monitoring method based on improved YOLOv8 model

Abstract

The invention discloses an engineering construction progress monitoring method based on an improved YOLOv8 model, and the method comprises the steps: dividing a photovoltaic base construction process into a plurality of stages, and obtaining a first aerial image of a key component of each stage of photovoltaic base construction to construct an image data set; the Invotion module is inserted into a detection head of the YOLOv8 model to obtain an improved YOLOv8 model; using the image data set to train the improved YOLOv8 model; obtaining a second aerial image of a built photovoltaic base, and analyzing the second aerial image by using the trained improved YOLOv8 model to obtain key components in the second aerial image; and determining a construction stage of the photovoltaic base under construction and calculating a construction progress of the construction stage based on key components in the second aerial image. The problems of high time consumption, high cost and low automation degree of traditional photovoltaic power generation base station construction progress monitoring are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of construction, and specifically relates to a method for monitoring the construction progress of a project based on an improved YOLOv8 model. Background Art

[0002] Under the background 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. The scale of photovoltaic power generation stations has been further expanded, and the number of construction projects will continue to increase. The management of construction progress has become increasingly complex and difficult, including problems such as a large number of participants, complex and changeable environments, and a large amount of data, resulting in low efficiency in construction progress management. Traditional monitoring of the construction progress of photovoltaic power generation stations requires the construction party to manually report the daily construction progress to the system, which is a labor-intensive task that is time-consuming, costly, and has a low degree of automation. Therefore, it is particularly important to achieve automated progress monitoring and improve the efficiency of progress management in complex environments. Summary of the Invention

[0003] In order to overcome the defects of the prior art, the present invention provides a method for monitoring the construction progress of a project based on an improved YOLOv8 model to solve the problems of time-consuming, high cost, and low automation in traditional monitoring of the construction progress of photovoltaic power generation stations.

[0004] To achieve the above object, the present invention provides a method for monitoring the construction progress of a project based on an improved YOLOv8 model, including the following steps:

[0005] Divide the construction process of the photovoltaic base into multiple stages, and obtain the first aerial images of the key components in each stage of the construction of the photovoltaic base to construct an image dataset;

[0006] Insert the Involution module into the detection head of the YOLOv8 model to obtain an improved YOLOv8 model;

[0007] Train the improved YOLOv8 model using the image dataset;

[0008] Obtain the second aerial image of the photovoltaic base under construction, and use the trained improved YOLOv8 model to analyze the second aerial image to obtain the key components in the second aerial image;

[0009] Based on the key components in the second aerial image, determine the construction stage of the photovoltaic base under construction and calculate the construction progress of the construction stage.

[0010] Furthermore, the construction process of the photovoltaic base is divided into four stages, namely the pile foundation construction stage, the photovoltaic support construction stage, and the photovoltaic panel laying stage.

[0011] Furthermore, the key components in the pile foundation construction stage are concrete pipe piles or steel spiral piles, the key components in the photovoltaic bracket construction stage are photovoltaic brackets, and the key components in the photovoltaic panel laying stage are photovoltaic panels.

[0012] Furthermore, when determining the construction stage of the photovoltaic base under construction, when the second aerial image only shows pile foundation components such as concrete pipe piles or steel spiral piles, the construction stage is determined to be the pile foundation construction stage; when the second aerial image includes pile foundation components and photovoltaic brackets, the construction stage is determined to be the photovoltaic bracket construction stage; when the second aerial image includes photovoltaic brackets and photovoltaic panels, the construction stage is determined to be the photovoltaic panel laying stage.

[0013] Furthermore, the detection head is a C2f module.

[0014] Furthermore, when implementing the step of training the improved YOLOv8 model using the image dataset, mAP@50 and mAP@50:95 are used as performance evaluation indicators of the improved YOLOv8 model.

[0015] The beneficial effects of the present invention are that the engineering construction progress monitoring method based on the improved YOLOv8 model of the present invention analyzes the photovoltaic power station construction process, improves the target detection model to detect key components in the photovoltaic power station construction process, and post-processes the detection data based on the improved YOLOv8 target detection model to obtain the progress information of the target plot according to the type and quantity of the detected components. The engineering construction progress monitoring method based on the improved YOLOv8 model of the present invention can not only solve the problem of small target detection during high-altitude flight of drones, but also achieve lightweight model, thereby realizing visualization, refinement, standardization and intelligence of photovoltaic power station management, effectively improving the construction quality and comprehensive benefits of photovoltaic power station projects, and is of great significance to promoting the digitalization and intelligent management of photovoltaic power station projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0017] Figure 1 The figure is a flow chart of a method for monitoring construction progress based on an improved YOLOv8 model according to an embodiment of the present invention.

[0018] Figure 2 Schematic diagram of the YOLOv8 model optimization process according to an embodiment of the present invention.

[0019] Figure 3 The following are the detection effects of different models in three specific scenarios. DETAILED DESCRIPTION

[0020] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention and are not intended to limit the invention. Additionally, it should be noted that for ease of description, only the parts related to the invention are shown in the drawings.

[0021] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below by referring to the drawings and in conjunction with the embodiments.

[0022] Referring to Figures 1 to 3 as shown, the present invention provides a method for monitoring the progress of engineering construction based on an improved YOLOv8 model, including the following steps:

[0023] S1. Divide the process of photovoltaic base construction into multiple stages, and obtain the first aerial images of the key components in each stage of photovoltaic base construction to construct an image dataset.

[0024] Divide the process of photovoltaic base construction, collect aerial images for the key components in each stage, and form a photovoltaic base progress recognition image dataset through instance annotation.

[0025] As a preferred implementation, the process of photovoltaic base construction is divided into four stages, namely the pile foundation construction stage, the photovoltaic support construction stage, and the photovoltaic panel laying stage.

[0026] In this embodiment, the key components in the pile foundation construction stage are concrete pipe piles or steel spiral piles, the key components in the photovoltaic support construction stage are photovoltaic supports, and the key components in the photovoltaic panel laying stage are photovoltaic panels.

[0027] Collect the aerial images corresponding to the above four stages through Internet open-source images and camera real-scene shooting. Use Labelme as the image annotation tool software, and use the Polygen tool to perform instance annotation on the key components in the collected images to determine the contours and positions of the key components, and form a photovoltaic base progress recognition image dataset.

[0028] Specifically, use the Polygen tool in the Labelme software to annotate the key components in the aerial images, and create a photovoltaic base progress recognition image dataset. This dataset contains a total of 10,555 annotated instances, including 4,233 concrete pile instances, 3,342 steel spiral pile instances, 1,935 photovoltaic support instances, and 1,045 photovoltaic panel instances.

[0029] S2. Insert the Involution module into the detection head of the YOLOv8 model to obtain an improved YOLOv8 model.

[0030] As a preferred embodiment, the detection head is a C2f module. This module adaptively analyzes the target and its surrounding environmental context through dynamically generated spatial-specific convolution kernels, enhances the feature representation of small target information, and solves the problem of difficult small target recognition. The YOLOv8 model is improved to enable the model to accurately recognize the key components in each construction stage in the case of target-background confusion and small targets.

[0031] This module extracts feature vectors from specific feature points of the image (small targets in the image) and processes them through a function that converts the feature vectors into a shape suitable as a kernel function. The transformed vectors form an Involution kernel customized for this feature point, which is used to process the surrounding feature vectors. Multiply this kernel by the pixels near the feature point to generate the Involution kernel result of K×K×1. This method uses the Involution module to dynamically generate specific cores adapted to different spatial positions according to the input features. By implementing differential spatial weighting, the feature representation of small target information is enhanced, thereby improving the accuracy of target analysis.

[0032] In view of the problems of target-background confusion and tiny target recognition in remote sensing image analysis, the YOLOv8 model is improved as follows:

[0033] The present invention proposes an innovative solution to the problem of target feature extraction limitation caused by the traditional target detection method's failure to fully utilize all pixel information in the image and thus ignoring rich context information.

[0034] The YOLOv8 model is improved by introducing a module called Involution module at the head of the detection network to enhance the target detection performance.

[0035] As Figure 2 shown, the specific optimization process for improving the YOLOv8 model is as follows:

[0036] Extract feature vectors from specific feature points of the image (such as the baseball field features of small objects) and process them through a function to convert them into a shape suitable as a kernel.

[0037] 1) Extract feature vectors from specific feature points of the image and process them through a function to convert them into a shape suitable as a kernel.

[0038] 2) After channel-to-space conversion, the vector is reshaped into a dedicated Involution kernel for this specific feature point and then applied to the neighboring feature vectors around this point.

[0039] 3) The product of this kernel and the pixels near the feature points generates the Involution kernel result of K×K×1.

[0040] 4) The same operation is performed on all channels to form a three-dimensional matrix of K×K×C.

[0041] 5) Finally, the features in the width and height dimensions are accumulated while keeping the features in the channel dimension unchanged.

[0042] Generally speaking, the introduced Involution module can dynamically generate kernel functions for specific spatial positions based on the input features and perform adaptive weighted processing. In this way, the Involution module dynamically creates specific kernels for different spatial positions based on the input features, realizes differential spatial weighting, thus significantly enhancing the feature expression of small target information and improving the accuracy of target analysis.

[0043] S3. Use the image dataset to train the improved YOLOv8 model.

[0044] Use the photovoltaic base progress recognition image dataset to train the improved YOLOv8 model, and apply the trained YOLOv8 model to perform target recognition on the aerial images of the photovoltaic base.

[0045] When implementing the step of using the image dataset to train the improved YOLOv8 model, mAP@50 and mAP@50:95 are used as the performance evaluation metrics of the improved YOLOv8 model.

[0046] To verify the effectiveness of the improvement of the YOLOv8 model, a series of ablation experiments are carried out for verification. These experiments divide the photovoltaic image dataset established in step S1 into a training set, a validation set and a test set according to the ratio of 7:2:1 to ensure a comprehensive evaluation of the performance of the model improvement.

[0047] The experimental hyperparameters are set as shown in Table 1.

[0048] Table 1. Ablation experiment parameter setting table

[0049]

[0050] The experimental environment is configured as follows: the operating system uses Ubuntu 18.04, the programming language is Python3.8.13, CUDA11.4 and cuDNN 8.2.2 are used for deep learning acceleration, and the deep learning framework is PyTorch 1.10.2.

[0051] The hardware platform is configured with an RTX-3090 GPU with 8GB of memory, and the processor is an Intel Core i7-6500M CPU with a main frequency of 3.20GHz.

[0052] To measure the effect of model improvement, the experiments were evaluated using the mAP@50 and mAP@50:95 metrics:

[0053] mAP is used to quantify the detection accuracy of the model at different confidence thresholds and is a standard metric for measuring the performance of object detection models.

[0054] The results of the ablation experiments are shown in Table 2, which presents the performance comparison of four models on two important evaluation metrics, mAP@50 and mAP@50:95. The experimental results show that the improved model, Photovoltaic Net, performs optimally on both metrics, reaching 0.840 and 0.391 respectively. The introduction of the Involution module can improve the algorithm performance by 3.7% (mAP@50). This achievement effectively combines the lightweight of small models and the high efficiency of large models, demonstrating a forward-looking application in photovoltaic detection technology.

[0055] Table 2. Results of ablation experiments

[0056]

[0057] As Figure 3 shown, it clearly demonstrates the detection effects of different models in three specific scenarios. From the results presented, the YOLOv7-tiny, YOLOv5s, and YOLOv8 models all exhibit certain problems of missed detection and false detection in these scenarios. These problems may affect the reliability and efficiency of the models in practical applications. In contrast, the PhotovoltaicNet model performs significantly better in each test scenario, significantly reducing the cases of missed detection and false detection.

[0058] S4. Obtain the second aerial image of the under-construction photovoltaic base and use the trained improved YOLOv8 model to analyze the second aerial image to obtain the key components in the second aerial image.

[0059] Obtain the analysis results of the second aerial image (such as an image or video) of the under-construction photovoltaic base, including the component categories and the number of components contained in a video.

[0060] S5. Based on the key components in the second aerial image, determine the construction stage of the under-construction photovoltaic base and calculate the construction progress of the construction stage.

[0061] In actual construction, the construction sequence in the same work area is as follows: first, pile foundations (concrete pipe piles or steel spiral piles) are laid, then photovoltaic supports are installed on the pile foundations, and finally photovoltaic panels are laid on the photovoltaic supports. Moreover, the second process will only start after the first process is completely finished. Therefore, only two types of components will be included in the same video, namely the type of component under current construction and the type of component corresponding to the previous process.

[0062] When determining the construction stage of the photovoltaic base under construction, when only pile foundation components such as concrete pipe piles or steel spiral piles are present in the second aerial image, the construction stage is determined to be the pile foundation construction stage. When the second aerial image includes pile foundation components and photovoltaic supports, the construction stage is determined to be the photovoltaic support construction stage. When the second aerial image includes photovoltaic supports and photovoltaic panels, the construction stage is determined to be the photovoltaic panel laying stage.

[0063] When calculating the construction progress of the construction stage, the construction progress is calculated using a formula. The formula is:

[0064] Construction progress = (Number of key components in the current stage / Known planned workload) × 100%.

[0065] The engineering construction progress monitoring method based on the improved YOLOv8 model of the present invention analyzes the construction process of the photovoltaic power station, improves the object detection model to detect key components in the construction process of the photovoltaic power station, post-processes the detection data based on the improved yolov8 object detection model, and obtains the progress information of the target plot according to the detected component types and quantities. The engineering construction progress monitoring method based on the improved YOLOv8 model of the present invention can not only solve the problem of small target detection during the high-altitude flight of the unmanned aerial vehicle, but also achieve the lightweight of the model (model pruning), thereby realizing the visualization, refinement, standardization, and intelligence of the management of the photovoltaic power station, effectively improving the construction quality and comprehensive benefits of the photovoltaic power station project, and having important significance for promoting the digital and intelligent management of the photovoltaic power station project.

[0066] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the technical solutions formed by mutually replacing the above features with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A construction progress monitoring method based on an improved YOLOv8 model, characterized in that: The following steps are involved: Dividing the photovoltaic base construction process into multiple stages, acquiring first aerial images of key components in each stage of the photovoltaic base construction to construct an image dataset; Insert the Involution module into the detection head of the YOLOv8 model to obtain an improved YOLOv8 model; Training the improved YOLOv8 model using the image dataset; Obtain a second aerial image of the photovoltaic base under construction, and use the trained improved YOLOv8 model to analyze the second aerial image to obtain key components in the second aerial image; Based on the key components in the second aerial image, the construction phase of the photovoltaic base under construction is determined and the construction progress of the construction phase is calculated.

2. The construction progress monitoring method based on the improved YOLOv8 model according to claim 1, wherein The photovoltaic base construction process is divided into four stages, namely the pile foundation construction stage, the photovoltaic support construction stage, and the photovoltaic panel laying stage.

3. The construction progress monitoring method based on the improved YOLOv8 model according to claim 2, wherein The key components in the pile foundation construction stage are concrete pipe piles or steel spiral piles, the key components in the photovoltaic bracket construction stage are photovoltaic brackets, and the key components in the photovoltaic panel laying stage are photovoltaic panels.

4. The construction progress monitoring method based on the improved YOLOv8 model according to claim 3 is characterized in that, When determining the construction stage of the photovoltaic base under construction, when the second aerial image only shows pile foundation components such as concrete pipe piles or steel spiral piles, the construction stage is determined to be the pile foundation construction stage; when the second aerial image includes pile foundation components and photovoltaic brackets, the construction stage is determined to be the photovoltaic bracket construction stage; when the second aerial image includes photovoltaic brackets and photovoltaic panels, the construction stage is determined to be the photovoltaic panel laying stage.

5. The construction progress monitoring method based on the improved YOLOv8 model according to claim 1, wherein The detection head is a C2f module.

6. The construction progress monitoring method based on the improved YOLOv8 model according to claim 1 is characterized in that, When implementing the step of training the improved YOLOv8 model using the image dataset, mAP@50 and mAP@50:95 are used as performance evaluation indicators of the improved YOLOv8 model.

Images