CN-121982482-A - Defect identification method and device for power grid equipment

CN121982482ACN 121982482 ACN121982482 ACN 121982482ACN-121982482-A

Abstract

The invention relates to a defect identification method and device of power grid equipment, and belongs to the technical field of image processing, wherein the defect identification method of the power grid equipment comprises the steps of acquiring a power grid equipment image acquired by an unmanned aerial vehicle, inputting the power grid equipment image into an image detection model to obtain an image detection result output by the image detection model, wherein the image detection model is based on a YOLO11 network, a multi-head space attention mechanism of a YOLO11 network back-bone part is replaced by a BiFormer attention mechanism, a convolution layer is AKConv, a feature fusion network of a YOLO11 network Neck part is replaced by an FFCA network, and the convolution layer is WTConv. The method and the device improve the image detection precision of the unmanned aerial vehicle power grid inspection and simultaneously reduce the calculation complexity of image detection.

Inventors

WU LINGRUI
LI YUTING
CHEN YUNFAN
YANG YIMING
YU TENGHUI

Assignees

湖北工业大学

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (10)

1. A method for identifying defects in power grid equipment, comprising: Acquiring a power grid equipment image acquired by an unmanned aerial vehicle; The method comprises the steps that an image of power grid equipment is input into an image detection model, an image detection result output by the image detection model is obtained, the image detection model is based on a YOLO11 network, a multi-head space attention mechanism of a YOLO11 network back part is replaced by a BiFormer attention mechanism, a convolution layer is AKConv, a feature fusion network of a YOLO11 network Neck part is replaced by an FFCA network, the convolution layer is WTConv, the image detection model takes a historical power grid equipment image as a sample during training, and a historical power grid equipment image carrying equipment labels and fault labels as labels.
2. The method of claim 1, wherein the attention weight matrix of BiFormer attention mechanism is determined based on the following formula: Wherein, the The attention weight matrix is represented as such, Representing the query matrix and, The matrix of keys is represented and, Is a scaling factor.
3. The method for identifying defects of power grid equipment according to claim 1, wherein the FFCA network comprises: The feature enhancement module is used for extracting multi-scale features based on multi-branch convolution; the feature fusion module is used for determining channel weights based on global average pooling and the multi-layer perceptron and carrying out channel weighted fusion; and the spatial context sensing module is used for carrying out feature extraction based on the spatial attention and the channel attention.
4. The method for identifying defects of a power grid device according to claim 1, wherein the AKConv convolution process comprises: initial sampling coordinate generation, dynamic offset learning and kernel feature extraction.
5. The method for identifying defects of a power grid device according to claim 1, wherein the WTConv convolution process comprises: Wavelet decomposition, multi-frequency domain convolution, and feature fusion.
6. The method of defect identification of electrical network equipment according to claim 4, further comprising: following the AKConv convolution process, a reparameterization acceleration is performed based on the channel reorganization and convolution operations.
7. The method for identifying defects in electrical network equipment according to claim 5, further comprising: After the WTConv convolution process, edge detection enhancement is performed based on the target box regression loss and the edge similarity loss.
8. A defect recognition device for power grid equipment, comprising: The acquisition module is used for acquiring the power grid equipment image acquired by the unmanned aerial vehicle; The detection module is used for inputting the power grid equipment image into the image detection model to obtain an image detection result output by the image detection model, the image detection model is based on a YOLO11 network, a multi-head space attention mechanism of a YOLO11 network back box part is replaced by a BiFormer attention mechanism, a convolution layer is AKConv, a feature fusion network of a YOLO11 network Neck part is replaced by an FFCA network, the convolution layer is WTConv, the image detection model takes a historical power grid equipment image as a sample during training, and a historical power grid equipment image carrying equipment labels and fault labels as labels.
9. An image detection apparatus comprising a memory and a processor, wherein, The memory is used for storing programs; The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps in the defect identification method of the power grid device according to any one of the preceding claims 1 to 7.
10. A computer readable storage medium storing a computer readable program or instructions which when executed by a processor is capable of carrying out the steps of the method for defect identification of a power grid device according to any one of the preceding claims 1 to 7.

Description

Defect identification method and device for power grid equipment Technical Field The present invention relates to the field of image processing technologies, and in particular, to a method and an apparatus for identifying defects of a power grid device. Background The rapid development of unmanned aerial vehicle technology provides a new solution for power grid inspection. The unmanned aerial vehicle can effectively improve the power grid inspection efficiency by virtue of flexible maneuverability and efficient covering capability. At present, commercial unmanned aerial vehicle is generally adopted in industry to acquire data, and equipment image data is acquired through visible light and infrared cameras. The existing intelligent inspection system for the power grid mainly has the defects that firstly, the performance of a traditional YOLO algorithm is obviously reduced under the scenes of complex illumination, target shielding, small-scale defect detection and the like, the detection rate of key defects such as insulator microcracks and pole tower corrosion is limited, and secondly, the existing model is high in calculation complexity, real-time reasoning is difficult to realize on mobile equipment, and on-site instant analysis and decision cannot be supported. Therefore, how to improve the image detection precision of the unmanned aerial vehicle power grid inspection system and reduce the calculation complexity of the image detection process becomes a technical problem to be solved urgently. Disclosure of Invention In view of the foregoing, it is necessary to provide a method and a device for identifying defects of a power grid device, which are used for solving the problems of lower image detection precision and higher calculation complexity in the existing inspection process of an unmanned aerial vehicle power grid. In order to solve the above problems, in a first aspect, the present invention provides a defect identifying method for a power grid device, including: Acquiring a power grid equipment image acquired by an unmanned aerial vehicle; The method comprises the steps that an image of power grid equipment is input into an image detection model, an image detection result output by the image detection model is obtained, the image detection model is based on a YOLO11 network, a multi-head space attention mechanism of a YOLO11 network back part is replaced by a BiFormer attention mechanism, a convolution layer is AKConv, a feature fusion network of a YOLO11 network Neck part is replaced by an FFCA network, the convolution layer is WTConv, the image detection model takes a historical power grid equipment image as a sample during training, and a historical power grid equipment image carrying equipment labels and fault labels as labels. In one possible implementation, the attention weight matrix of the BiFormer attention mechanism is determined based on the following formula: Wherein, the The attention weight matrix is represented as such,Representing the query matrix and,The matrix of keys is represented and,Is a scaling factor. In one possible implementation, the FFCA network includes: The feature enhancement module is used for extracting multi-scale features based on multi-branch convolution; the feature fusion module is used for determining channel weights based on global average pooling and the multi-layer perceptron and carrying out channel weighted fusion; and the spatial context sensing module is used for carrying out feature extraction based on the spatial attention and the channel attention. In one possible implementation, the AKConv convolution process includes: initial sampling coordinate generation, dynamic offset learning and kernel feature extraction. In one possible implementation, the WTConv convolution process includes: Wavelet decomposition, multi-frequency domain convolution, and feature fusion. In one possible implementation, the method further includes: following the AKConv convolution process, a reparameterization acceleration is performed based on the channel reorganization and convolution operations. In one possible implementation, the method further includes: After the WTConv convolution process, edge detection enhancement is performed based on the target box regression loss and the edge similarity loss. On the other hand, the invention also provides a defect identification device of the power grid equipment, which comprises the following components: The acquisition module is used for acquiring the power grid equipment image acquired by the unmanned aerial vehicle; The detection module is used for inputting the power grid equipment image into the image detection model to obtain an image detection result output by the image detection model, the image detection model is based on a YOLO11 network, a multi-head space attention mechanism of a YOLO11 network back box part is replaced by a BiFormer attention mechanism, a convolution layer is AKConv, a feature fusion network of a YOLO1