Power transmission line foreign matter identification system and method based on edge calculation and deep learning
By using edge computing and deep learning technology in the transmission line foreign object recognition system, the improved YOLOv7 network is trained and image enhancement processing is performed, which solves the problems of poor target detection capabilities and great environmental impact in the prior art, and achieves efficient and accurate foreign object recognition.
Patent Information
- Application Number
- CN202411656545.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-03-07
AI Technical Summary
The existing foreign object recognition technology in transmission lines has poor target detection capabilities and is greatly affected by the environment, resulting in a low recognition accuracy.
A transmission line foreign object recognition system based on edge computing and deep learning is adopted to train the improved YOLOv7 network through the cloud intelligent platform and send it to the edge computing terminal for image enhancement processing and foreign object recognition.
It improves the accuracy and reliability of foreign object recognition in transmission lines, solves the problems of poor target detection capabilities and great environmental impact, and achieves efficient identification in complex environments.
Smart Images
Figure CN119579977A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of foreign object recognition in power transmission lines, and in particular to a foreign object recognition system and method for power transmission lines based on edge computing and deep learning. Background Art
[0002] Due to the complex environment of power transmission lines, foreign objects such as kites and mechanical construction have become an important cause of line failure. In recent years, the informationization and intelligent operation and maintenance of power transmission lines have been continuously improved, and some key transmission lines have basically completed the installation of online monitoring equipment, which can effectively monitor the intrusion of foreign objects. However, the recognition and analysis of massive intrusion monitoring images have problems such as low accuracy.
[0003] In recent years, deep learning has been applied to the monitoring and identification of foreign objects on power transmission lines, and has achieved certain results. For example, image recognition technology is used to automatically identify wildfires on power transmission lines; foreign object detection is achieved by combining multi-network feature fusion with random forests, and a foreign object recognition model for power transmission lines in complex scenarios based on an improved YOLOv4 network is used. However, existing methods have poor target detection capabilities and are greatly affected by the environment. Summary of the invention
[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a transmission line foreign object identification system and method based on edge computing and deep learning, which solves the problems that the prior methods have poor target detection capabilities and are greatly affected by the environment.
[0005] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a transmission line foreign object identification system based on edge computing and deep learning, including an image acquisition device, an edge computing terminal and a cloud intelligent platform, the edge computing terminal is connected to the image acquisition device and the cloud intelligent platform respectively, the image acquisition device is used to collect transmission line inspection images, the edge computing terminal is used to process the collected image data, and the cloud intelligent platform uses the collected image data to train an improved network and send the trained network to the edge computing terminal.
[0006] The beneficial effect of the above scheme is: the power transmission line foreign object identification system proposed in the present invention mainly includes three parts: image acquisition equipment, edge computing terminal and cloud intelligent platform. The improved network is trained on the cloud intelligent platform and sent to the edge computing terminal. The collected inspection images are enhanced and processed on the edge computing terminal, and the improved YOLOv7 network is used to identify foreign objects to output the type of foreign objects in the power transmission line. The present invention solves the problem that the existing methods have poor target detection capabilities and are greatly affected by the environment.
[0007] In addition, the present invention also adopts a technical solution: a method for identifying foreign objects in power transmission lines based on edge computing and deep learning, comprising the following steps:
[0008] S1: Collecting transmission line inspection image data using image acquisition equipment;
[0009] S2: Use the cloud intelligence platform to train the improved YOLOv7 network and send the trained YOLOv7 network to the edge computing terminal;
[0010] S3: Use the edge computing terminal to perform image enhancement processing on the collected transmission line inspection image data, and use the improved YOLOv7 network to identify foreign objects in the enhanced image, obtain the type of foreign objects in the transmission line, and complete the transmission line foreign object identification based on edge computing and deep learning.
[0011] The beneficial effects of the above scheme are: in response to the problems of poor small target detection ability and large environmental impact in most existing methods, a transmission line foreign object recognition method based on edge computing and deep learning is proposed. Since the traditional centralized information processing mode is inefficient and does not conform to the wide range of transmission lines, the proposed method designs a transmission line foreign object recognition system based on edge computing to quickly process massive data nearby.
[0012] Furthermore, S2 uses polarized self-attention mechanism and Wise-IoU loss function to optimize YOLOv7.
[0013] The beneficial effect of the above further scheme is: in order to fully extract image features, the proposed method adopts the polarized self-attention (PSA) mechanism to improve the YOLOv7 network for dual-channel feature extraction, and at the same time uses the Wise-IoU loss to improve the generalization ability of the network model, further improving the reliability of foreign object recognition.
[0014] Furthermore, the polarized self-attention mechanism adopts a parallel mode.
[0015] The beneficial effect of the above further scheme is: through the above technical scheme, the model is helped to capture important contextual information in the image, improve sensitivity to the target, focus on the target area, reduce background interference, and make the model more robust and effective when processing images with complex backgrounds and multiple targets.
[0016] Furthermore, the Wise-IoU loss function L WIoU for:
[0017]
[0018] Among them, xba and ba is the center point coordinate of the anchor box, x ot and ot is the center point coordinate of the target box, * indicates the calculation graph separation operation, W i and H i is the width and height of the overlapping area, L IoU is the first parameter.
[0019] The beneficial effects of the above further scheme are: Wise-IoU constructs distance attention through distance metric, has a two-layer attention mechanism, and solves the unreasonable design problem of CIoU adding aspect ratio into the loss function. Wise-IoU weakens the penalty of geometric factors by separating the width and height of the minimum bounding box, achieves less training intervention, and improves the generalization ability of the model.
[0020] Furthermore, the Mosaic-9 method is used in S3 to perform image enhancement processing.
[0021] The beneficial effects of the above further scheme are: Mosaic enhancement is a widely used image processing technology that improves image quality by dividing an image into multiple blocks or synthesizing multiple images into one image, thereby enhancing the robustness and accuracy of the network. Since transmission line inspection images usually contain a large amount of background, feature information and targets of various sizes, this places high demands on the generalization ability of the algorithm and the diversity of the features and background information of the input images during training. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 This is a diagram of the transmission line foreign object recognition system based on edge computing and deep learning.
[0023] Figure 2 This is a flow chart of the foreign object identification method for transmission lines based on edge computing and deep learning.
[0024] Figure 3 This is the network structure diagram of the parallel polarized self-attention mechanism.
[0025] Figure 4 It is the loss function curve.
[0026] Figure 5 The following are the results of identifying different foreign objects.
[0027] Figure 6 Identify result graphs for different environments. DETAILED DESCRIPTION
[0028] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.
[0029] Embodiment 1, as Figure 1As shown, a transmission line foreign object recognition system based on edge computing and deep learning includes an image acquisition device, an edge computing terminal and a cloud intelligent platform. The edge computing terminal is connected to the image acquisition device and the cloud intelligent platform respectively. The image acquisition device is used to acquire transmission line inspection images. The edge computing terminal is used to process the acquired image data. The cloud intelligent platform uses the acquired image data to train an improved network and sends the trained network to the edge computing terminal.
[0030] In one embodiment of the present invention, the cloud intelligence platform trains the improved network based on hardware resources, and the hardware resources include an operating environment, an image model library, and an image sample library; the edge computing terminal mainly includes an edge intelligence chip, an embedded operating system, and an edge intelligence framework, which mainly realizes image storage, lightweight model establishment, and sample transmission.
[0031] Embodiment 2, as Figure 2 As shown, a method for identifying foreign objects in a power transmission line based on edge computing and deep learning includes the following steps:
[0032] S1: Collecting transmission line inspection image data using image acquisition equipment;
[0033] S2: Use the cloud intelligence platform to train the improved YOLOv7 network and send the trained YOLOv7 network to the edge computing terminal;
[0034] S3: Use the edge computing terminal to perform image enhancement processing on the collected transmission line inspection image data, and use the improved YOLOv7 network to identify foreign objects in the enhanced image, obtain the type of foreign objects in the transmission line, and complete the transmission line foreign object identification based on edge computing and deep learning.
[0035] In the process of foreign body identification on power transmission lines, firstly, the inspection images of power transmission lines are acquired online using cameras and other acquisition devices, and Mosaic-9 enhancement is performed. Then, the improved YOLOv7 network is trained using the image set on the cloud intelligence platform, and the trained network is sent to the edge computing terminal. Finally, on the edge computing terminal, the trained improved YOLOv7 network is used to identify foreign bodies on the power transmission lines, and the type of foreign bodies is output, and the images are transmitted to the cloud intelligence platform to update its image set.
[0036] The traditional YOLOv7 design uses the Extended Efficient Aggregation Network (E-ELAN) to enhance the network learning ability and improve the efficiency of feature extraction. The MP module is used for downsampling, and the traditional pooling and convolution are fused to reduce the loss of features in the process. In the backbone network (Backbone), the SPP module (SPPCSPC) is improved, and convolution is introduced between pooling operations to avoid problems such as feature loss, and a feature map with richer features is output. In the neck network (Neck), the PANet structure is used for network aggregation, so that different feature maps can be effectively fused. In the prediction head network (Head), REP-Conv is used to adjust the number of channels. YOLOv7 parses and filters the detection results by applying anchor boxes and non-maximum suppression.
[0037] The PSA mechanism is a more sophisticated dual attention mechanism. The introduction of PSA enables the YOLOv7 model to focus on the target-related image area and improve the accuracy of target recognition. For the complex background of power transmission lines, PSA helps the model capture important contextual information in the image, improves sensitivity to the target, and focuses on the target area, reducing background interference, making the model more robust and effective when processing images with complex backgrounds and multiple targets.
[0038] PSA has two branches: the self-attention mechanism of the channel and spatial dimensions, and the results of the two branches are fused to obtain the output of the structure. PSA has two modes: series and parallel. The proposed method adopts the parallel type.
[0039] In S2, polarized self-attention mechanism and Wise-IoU loss function are used to optimize YOLOv7.
[0040] The polarized self-attention mechanism adopts a parallel mode, such as Figure 3 shown.
[0041] Wise-IoU loss function L WIoU for:
[0042]
[0043] Among them, x ba and ba is the center point coordinate of the anchor box, x ot and ot is the center point coordinate of the target box, * indicates the calculation graph separation operation, W i and H i is the width and height of the overlapping area, L IoU is the first parameter.
[0044] The Mosaic-9 method is used for image enhancement in S3.
[0045] Before inputting the foreign object recognition model, 9 images from the input data are randomly selected and different enhancement methods, such as cropping and rotation, are used, and they are spliced into one image and then input into the network. The image spliced by the Mosaic-9 enhancement method contains more feature information, background information, and objects of different scales, which enhances the diversity feature information and background information of the data samples, and can effectively improve the generalization ability of the network and reduce model overfitting.
[0046] In one embodiment of the present invention, the data set used in the experiment includes real-life 10kV, 110kV, 220kV, 500kV AC transmission lines and ±800kV, ±1000kV UHV AC and DC transmission lines. Among them, foreign objects in the transmission lines include four categories: kites, bird nests, garbage (plastic bags, etc.), and mechanical construction, containing a total of 25,760 images. In the experiment, the hardware environment of the cloud intelligence platform is the Windows 11 operating system, using the deep learning framework Pytorch 3.8.13; the hardware environment of the edge computing terminal is the CPU Nvidia Carmeal-ARM v8.2 6C6T-6MB L2+4MB L3, using the deep learning framework Pytorch 3.8.10. In the experimental parameters, the Batch Size is 64 and the number of iterations is 200.
[0047] Comparing the YOLOv7 network and the improved YOLOv7 network model, the total loss is as follows Figure 4 As shown. Figure 4 It can be seen that there is no obvious overfitting phenomenon in the training of the improved YOLOv7 model, and the experimental results are reasonable. The total loss of the YOLOv7 network is stable at 0.25, and the total loss of the improved YOLOv7 network is stable at 0.11. In addition, the improved YOLOv7 network converges faster because it uses PSA and Wise-IoU loss functions, which can improve the model recognition performance while converging quickly, and the bounding box regression is more stable.
[0048] In order to verify the effectiveness of the proposed method in the actual transmission line scene, the image set is identified, and the recognition effects of four foreign objects such as bird nests, garbage, kites, and mechanical construction are shown as follows: Figure 5 shown.
[0049] Depend on Figure 5It can be seen that the proposed method can accurately identify various foreign objects, even bird nests that are far away can be reliably identified. To verify the superiority of the proposed method, it is compared with the prior art 1 and the prior art 2. The recognition accuracy of the three methods for various foreign objects on the transmission line is shown in Table 1.
[0050] Table 1 Recognition accuracy of different methods
[0051]
[0052] As shown in Table 1, the proposed method achieves ideal results by enhancing images and using the improved YOLOv7 network to identify foreign objects on power transmission lines, with an average recognition accuracy of 92.73%. In particular, the recognition accuracy of mechanical construction exceeds 90%, and its image may be misplaced, resulting in misjudgment of crossing the line, but the proposed method can reliably identify it. Comparative Document 1 uses an improved YOLOv4 network to identify foreign objects on power transmission lines, but the network has poor recognition effect on small targets such as garbage, only 87.38%, and there are certain misjudgments for mechanical construction. Comparative Document 2 uses an improved ResNet-18 neural network to realize the recognition of foreign objects on power transmission lines, especially bird nests, with an accuracy of 91.17%, but the recognition accuracy of other types of foreign objects is not high, with an average of only 87.76%, which is 4.97% lower than that of the proposed method.
[0053] The environment of power transmission lines is complex, and the vision is good in clear weather. The foreign object recognition effect of the proposed method is very ideal. In order to demonstrate the universality of the proposed method, transmission line images on rainy and foggy days were selected for testing. The results are as follows Figure 6 And as shown in Table 2.
[0054] Table 2 Recognition accuracy in different environments
[0055]
[0056] Depend on Figure 6 It can be seen that even if the image is blurred, the proposed method can still effectively identify foreign objects on the transmission line in foggy and rainy days. Combined with Table 2, the average accuracy of its recognition of various foreign objects is about 89%, which is about 3% lower than the recognition accuracy on sunny days. Because in foggy or rainy environments, garbage such as transparent plastic bags are similar to the environment and are easy to misjudge, the recognition accuracy drops significantly. But overall, the proposed method is effective for foreign object recognition on transmission lines in complex environments.
[0057] The present invention proposes a transmission line foreign body identification system and method based on edge computing and deep learning. In the transmission line foreign body identification system based on edge computing, the cloud intelligent platform is used to train the improved YOLOv7 network and send it to the edge computing terminal. The edge computing terminal uses the trained improved YOLOv7 network to identify foreign bodies after enhancing the collected inspection images, and obtains the type of foreign bodies in the transmission line. The experimental results based on the Pytorch platform show that:
[0058] (1) The improved YOLOv7 network adopts PSA and Wise-IoU loss functions to improve the recognition performance of the model, and its average recognition accuracy reaches 92.73%.
[0059] (2) The proposed method updates the image set and improves the YOLOv7 network model in real time through cloud-edge collaboration. Therefore, the foreign object recognition effect in complex environments is ideal. The average accuracy of various foreign object recognition is about 89%, which is about 3% lower than the recognition accuracy on sunny days.
[0060] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.
Claims
1. A power transmission line foreign object recognition system based on edge computing and deep learning, characterized in that: It includes an image acquisition device, an edge computing terminal and a cloud intelligent platform. The edge computing terminal is connected to the image acquisition device and the cloud intelligent platform respectively. The image acquisition device is used to acquire transmission line inspection images. The edge computing terminal is used to process the acquired image data. The cloud intelligent platform uses the acquired image data to train an improved network and sends the trained network to the edge computing terminal.
2. A method for identifying foreign objects in power transmission lines based on edge computing and deep learning, characterized in that: The following steps are involved: S1: Collecting transmission line inspection image data using image acquisition equipment; S2: Use the cloud intelligence platform to train the improved YOLOv7 network and send the trained YOLOv7 network to the edge computing terminal; S3: Use the edge computing terminal to perform image enhancement processing on the collected transmission line inspection image data, and use the improved YOLOv7 network to identify foreign objects in the enhanced image, obtain the type of foreign objects in the transmission line, and complete the transmission line foreign object identification based on edge computing and deep learning.
3. The method for identifying foreign objects in power transmission lines based on edge computing and deep learning according to claim 2 is characterized in that: In S2, the polarized self-attention mechanism and the Wise-IoU loss function are used to optimize YOLOv7.
4. The method for identifying foreign objects in power transmission lines based on edge computing and deep learning according to claim 3 is characterized in that: The polarized self-attention mechanism adopts a parallel mode.
5. The method for identifying foreign objects in power transmission lines based on edge computing and deep learning according to claim 3 is characterized in that: The Wise-IoU loss function L WIoU for: Among them, x ba and ba is the center point coordinate of the anchor box, x ot and ot is the center point coordinate of the target box, * indicates the calculation graph separation operation, W i and H i is the width and height of the overlapping area, L IoU is the first parameter.
6. The method for identifying foreign objects in power transmission lines based on edge computing and deep learning according to claim 2, characterized in that: In the S3, the Mosaic-9 method is used to perform image enhancement processing.