Electrical machinery construction hidden danger identification method, equipment and medium
By using an improved YOLOv8 model, combined with CFDCN, MSPAM, and PCIoU, efficient and real-time identification of potential hazards in power machinery construction was achieved, solving the problem of low efficiency in manual identification and improving identification efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-04-10
AI Technical Summary
In current power machinery construction, manual hazard identification is inefficient and lacks real-time capability, making it difficult to meet the real-time risk management needs of power construction.
An improved YOLOv8 model is adopted, which combines a center-fixed deformable convolutional network (CFDCN), a multi-scale perceptual attention module (MSPAM), and a proportional consistency loss function (PCIoU) for identifying potential hazards in power machinery construction. End-to-end target detection is achieved through machine vision technology.
It improves the efficiency and time coverage of identifying potential hazards in power machinery construction, meets real-time requirements, and enhances identification accuracy and training convergence speed.
Smart Images
Figure CN121837595A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power construction, and in particular to a method, equipment and medium for identifying potential hazards in power machinery construction. Background Technology
[0002] Power engineering is a core component of the national energy infrastructure. With the accelerated construction of new power systems and the large-scale advancement of ultra-high voltage (UHV) and smart grid projects, power construction is exhibiting new characteristics: high voltage levels, large operational spans, and complex construction environments. The types and quantities of power machinery and equipment used in construction, such as tower cranes, crawler cranes, and excavators, have increased significantly, posing safety hazards to power engineering projects. Especially in complex construction environments, improper operation of power machinery and equipment can lead to safety accidents such as mechanical collisions and insufficient safety distances, resulting in personal injury and economic losses.
[0003] Currently, the main method for identifying potential hazards in power machinery construction is a combination of manual inspection and traditional monitoring. This relies on on-site visual inspections by technicians and measurements using handheld instruments. Limited by personnel experience and inspection cycles, this approach struggles to cover the entire operational period of the power machinery. Furthermore, some data processing is done manually, which is insufficient to meet the real-time risk management needs of power construction.
[0004] Therefore, utilizing intelligent methods for identifying potential hazards in power machinery construction and reducing the impact of manual operations has become crucial for ensuring the safety of power engineering construction. Summary of the Invention
[0005] The purpose of this invention is to propose a method, equipment, and medium for identifying potential hazards in power machinery construction, thereby solving the technical problems of low efficiency and insufficient real-time performance in manual hazard identification.
[0006] Specifically, the present invention provides a method for identifying potential hazards in the construction of power machinery, comprising the following steps:
[0007] S1. Obtain a dataset of images from power construction sites;
[0008] S2. Construct an improved YOLOv8 model; the improved YOLOv8 model introduces a center-fixed deformable convolutional network CFDCN, a multi-scale perceptual attention module MSPAM, and a proportional consistency loss function PCIoU (Proportional Consistency Intersection over Union) on the basis of the YOLOv8 model.
[0009] S3. Use the power construction site image dataset to train the improved YOLOv8 model to obtain the trained model;
[0010] S4. Use the trained model to identify potential hazards at the power construction site.
[0011] Furthermore, in step S1, when obtaining the power construction site image dataset, Labelme software is used for manual labeling.
[0012] Furthermore, the center-fixed deformable convolutional network CFDCN is used in the feature extraction network of the YOLOv8 model to enhance the ability to extract features of foldable components of electric machinery. CFDCN is used to replace the convolution calculation of the C2f module in the YOLOv8 model. The foldable components include arms and buckets.
[0013] Furthermore, the Multi-Scale Perception Attention (MSPAM) module is added after the SPPF module of the YOLOv8 model to enhance the ability to recognize electromechanical targets of different sizes in the image.
[0014] Furthermore, the enhanced feature map obtained through the multi-scale perceptual attention module MSPAM is as follows:
[0015]
[0016] Where F is the input feature map, This is the feature map after channel convolution compression. Let k be a convolution function with a kernel size of k×k. This is the activation function.
[0017] Furthermore, during the training process in step S3, the improved PCIoU loss function is used. The PCIoU loss function is based on arm features and is specifically expressed as follows:
[0018]
[0019] in, To predict the intersection ratio of the bounding box area with the actual electric machinery and arm, ; To ensure consistency between the predicted and actual electric machinery and the arm center, , To ensure the prediction is consistent with the actual electric machinery and boom ratio, S 预测车 S 真实车 S 预测臂 S 真实臂P represents the area of the predicted electric machinery, the actual electric machinery, the predicted arm, and the actual arm, respectively; 预测臂中心 P 真实臂中心 P 预测车中心 P 真实车中心 These represent the coordinates of the predicted arm center, the actual arm center, the predicted power machinery center, and the actual power machinery center, respectively. A non-transitory computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the aforementioned method for identifying potential hazards in power machinery construction.
[0020] A device for identifying potential hazards during power machinery construction includes: a processor, a storage device, and a computer program stored in the storage device and executable on the processor. When the processor executes the computer program, it implements the method for identifying potential hazards during power machinery construction.
[0021] The beneficial effects of this invention are as follows: This invention utilizes machine vision technology to replace manual safety hazard identification in power machinery construction, improving identification efficiency and time coverage. Based on the efficient single-stage architecture of the YOLO series machine vision object detection model YOLOv8, it achieves end-to-end direct prediction of power construction machinery categories and bounding boxes, meeting the requirements of real-time identification. Simultaneously, to improve the accuracy of identifying safety hazards in power construction machinery, this invention introduces a center-fixed deformable convolutional network (CFDCN) to enhance feature extraction from the arm portion of the power machinery, introduces a multi-scale perceptual attention module (MSPAM) to improve the perception capability of power machinery of different sizes in machine vision, and introduces a proportional consistency loss function (PCIoU) to optimize the training convergence speed of the model. Attached Figure Description
[0022] Figure 1 This is a flowchart of a method for identifying potential hazards in power machinery construction according to an embodiment of the present invention;
[0023] Figure 2 This is a schematic diagram of the overall structure of the model of the present invention;
[0024] Figure 3 This is a diagram showing the experimental results of the present invention;
[0025] Figure 4 This is a structural schematic diagram of a power machinery construction hazard identification device according to an embodiment of the present invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.
[0027] Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.
[0028] Please refer to Figures 1-2 , Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the model structure.
[0029] This invention provides a method for identifying potential hazards during the construction of electrical machinery, comprising the following steps:
[0030] S1. Obtain the image dataset of power construction machinery;
[0031] Images of power construction machinery were collected through the internet and captured from images taken at construction sites. The internet-collected images primarily included static appearance and dynamic operational images of various common power construction machinery such as cranes, excavators, pole erection machines, and cable laying machines. The construction site images were obtained through high-definition surveillance videos deployed at key locations within the construction area, or by capturing images of power machinery in operation taken by on-site personnel.
[0032] Then, the Labelme software was used for manual labeling. Polygons were used to mark the overall structure and key components (such as arms, rods, hooks, etc.) of the power machinery in the collected images. The images and labels were saved respectively to form a power construction machinery image dataset.
[0033] S2. Construct an improved YOLOv8 model; the improved YOLOv8 model introduces a center-fixed deformable convolutional network CFDCN and a multi-scale perceptual attention module (MSPAM) on the basis of the YOLOv8 model.
[0034] It should be noted that the Center-Fixed Deformable Convolutional Network (CFDCN) is introduced into the neck part of the YOLOv8 model to enhance the extraction capability of features from foldable components such as the boom and bucket of electromechanical equipment. CFDCN replaces the convolution calculation in the C2f module of the YOLOv8 model. CFDCN uses a 3×3 deformable mesh for sampling, followed by weighted summation. The 3×3 deformable mesh is obtained by superimposing the offset Δ of the mesh points at the positions R of a traditional 3×3 mesh, thus deforming the mesh shape. The offset of the center point of the 3×3 mesh is fixed at 0 to ensure that the convolution calculation covers the entire range. The formula for calculating the convolution of points is:
[0035]
[0036] in, w is the weight of the corresponding point. This represents the offset of the grid point.
[0037] It should be noted that the Multi-Scale Perception Attention Module (MSPAM) is used after the SPPF module of the YOLOv8 model to enhance the ability to recognize electromechanical targets of different sizes in the image.
[0038] The MSPAM computational process includes: channel convolutional compression, spatial multi-scale feature extraction, activation function processing, and output via residual connections. Channel convolutional compression performs convolutional calculations along the channel dimension, outputting a feature map with a channel dimension of 1, thus compressing the channel dimension. Spatial multi-scale feature extraction involves performing convolutional calculations in the compressed feature map using three different kernel sizes (3×3, 5×5, and 7×7) to extract features from receptive fields of different sizes, enhancing the ability to extract targets at different scales. The convolutional calculation results are processed using the ReLU activation function and then multiplied with the output. Finally, the enhanced feature map is obtained by outputting via residual connections.
[0039]
[0040] Where F is the input feature map, This is the feature map after channel convolution compression. Let k be a convolution function with a kernel size of k×k. This is the activation function.
[0041] S3. Use the power construction site image dataset to train the improved YOLOv8 model to obtain the trained model;
[0042] It should be noted that during the training process in step S3, the improved PCIoU loss function is used. The PCIoU loss function is based on arm features and is specifically expressed as follows:
[0043]
[0044] in, To predict the intersection ratio of the bounding box area with the actual electric machinery and arm, ; To ensure consistency between the predicted and actual electric machinery and the arm center, , To ensure the prediction is consistent with the actual electric machinery and boom ratio, S 预测车 S 真实车 S 预测臂 S 真实臂 P represents the area of the predicted electric machinery, the actual electric machinery, the predicted arm, and the actual arm, respectively; 预测臂中心 P 真实臂中心 P 预测车中心 P 真实车中心These represent the coordinates of the predicted arm center, the actual arm center, the predicted electromechanical center, and the actual electromechanical center, respectively.
[0045] In this invention, central consistency is achieved. As a penalty term, the distance between the center points of the predicted bounding box and the ground truth bounding box can be directly optimized, accelerating the convergence speed. By maintaining the area of the most characteristic arm in the electromechanical equipment consistent with the overall area, a principle of proportional consistency for electromechanical equipment is proposed. As a penalty. When When the area of overlap between the predicted and ground truth bounding boxes is small and the distance between them is large, it can provide additional geometric constraints for the regression training, allowing the network to more easily learn how to adjust the position for boosting. At the same time, when When the gradient approaches zero, direct optimization leads to gradient vanishing. Therefore, we introduce electromechanical ratio consistency. Afterwards, additional gradients can be provided to make training more stable.
[0046] S4. Use the trained model to identify potential hazards at the power construction site.
[0047] Please refer to Figure 3 and Table 1, Figure 3 This is a schematic diagram of the measured results of the present invention. Table 1 is a table of the results of the present invention.
[0048] Table 1
[0049]
[0050] The trained model is applied to construction site images. After identifying electrical machinery, a spatial ranging algorithm can determine the distance between the electrical machinery and other equipment. This allows for real-time monitoring of potential hazards at the power construction site, in accordance with operational specifications.
[0051] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a power machinery construction hazard identification device 401 according to an embodiment of the present invention. The power machinery construction hazard identification device 401 includes a processor 402, a storage device 403, and a computer program stored in the storage device 403 and executable on the processor 402. When the processor 402 executes the computer program, it implements the power machinery construction hazard identification method.
[0052] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for identifying potential hazards in power machinery construction.
[0053] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for identifying potential hazards in power machinery construction, characterized in that: Includes the following steps: S1. Obtain a dataset of images from power construction sites; S2. Construct an improved YOLOv8 model; the improved YOLOv8 model introduces a center-fixed deformable convolutional network CFDCN, a multi-scale perceptual attention module MSPAM, and a proportional consistency loss function PCIoU on the basis of the YOLOv8 model. S3. Use the power construction site image dataset to train the improved YOLOv8 model to obtain the trained model; S4. Use the trained model to identify potential hazards at the power construction site.
2. The method for identifying potential hazards in power machinery construction as described in claim 1, characterized in that: In step S1, when obtaining the power construction site image dataset, Labelme software was used for manual labeling.
3. The method for identifying potential hazards in power machinery construction as described in claim 1, characterized in that: The center-fixed deformable convolutional network CFDCN is used in the feature extraction network of the YOLOv8 model to enhance the ability to extract features of foldable components of electric machinery. CFDCN is used to replace the convolution calculation of the C2f module in the YOLOv8 model. The foldable components include arms and buckets.
4. The method for identifying potential hazards in power machinery construction as described in claim 1, characterized in that: The Multi-Scale Perception Attention (MSPAM) module is added after the SPPF module of the YOLOv8 model to enhance the ability to recognize electromechanical targets of different sizes in the image.
5. The method for identifying potential hazards in power machinery construction as described in claim 4, characterized in that: The enhanced feature map obtained by the Multi-Scale Perception Attention Module (MSPAM) is as follows: ； Where F is the input feature map, This is the feature map after channel convolution compression. Let k be a convolution function with a kernel size of k×k. This is the activation function.
6. The method for identifying potential hazards in power machinery construction as described in claim 1, characterized in that: During training in step S3, the improved PCIoU loss function is used. The PCIoU loss function is based on arm features and is specifically expressed as follows: ； in, To predict the intersection ratio of the bounding box area with the actual electric machinery and arm, ; To ensure consistency between the predicted and actual electric machinery and the arm center, , To ensure the prediction is consistent with the actual electric machinery and boom ratio, S 预测车 S 真实车 S 预测臂 S 真实臂 P represents the area of the predicted electric machinery, the actual electric machinery, the predicted arm, and the actual arm, respectively; 预测臂中心 P 真实臂中心 P 预测车中心 P 真实车中心 These represent the coordinates of the predicted arm center, the actual arm center, the predicted electromechanical center, and the actual electromechanical center, respectively.
7. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for identifying potential hazards in power machinery construction as described in any one of claims 1 to 6.
8. A device for identifying potential hazards in power machinery construction, characterized in that: include: A processor, a storage device, and a computer program stored on the storage device and executable on the processor, wherein the processor executes the computer program to implement the method for identifying potential hazards in power machinery construction as described in any one of claims 1 to 6.