Insulator defect detection for power grid based on light correction enhancement and YOLOv5 model. (November 2022)

• Record Type: Journal Article
• Authors: Li, Yi; Ni, Minzhe; Lu, Yanfeng
• Is Part Of: Energy reports. Volume 8(2022)Supplement 13
• Page Start: 807
• Page End: 814
• Publication Date: 2022-11
• Subjects: Defect detection -- Power grid diagnosis -- Vision detection; Power resources -- Periodicals; Energy industries -- Periodicals; Power resources; Periodicals; Electronic journals; 621.04205
• DOI: 10.1016/j.egyr.2022.08.027
• Languages: English
• ISSNs: 2352-4847

Abstract

Abstract: To guarantee the safety of the power grid system, it is essential to proceed reliable powerline inspection. Insulators are key devices in the powerlines. Their major function is to achieve mechanical fixing and electrical insulation, they play a key role in power lines. Insulators are deployed outdoors. Therefore, ensuring the safe operation of insulators is significant in the powerline inspection. Among all the inspection method, visual inspection is the key way. However, problems such as large changes in outdoor lighting have a strong impact on the accuracy of insulator detection. To overcome the shortcomings of uneven illumination, low contrast and poor details display in outdoor images, in this paper we introduces an image enhancement method based on illumination correction and compensation. First, the input data is converted from RGB color space to the HSV space, and three components, H, S and V, are obtained. The saturation component S is enhanced adaptively, and the brightness component V is processed by multi-scale gradient domain guided filter (MGDGF). Then the illumination component of the image is extracted, and corrected by two-dimensional adaptive Gamma transformation. The new brightness component is fused by Retinex based models. It helps to enhance the dark details and overall brightness of the image. This method not only solves the uneven illumination problem of the image, but also improves the contrast and details, while maintaining the original naturalness. Further, we introduce a real-time one step detection model based on YOLOv5, to detect the defect of the insulator. We evaluate the proposed method on an open public dataset. The evaluation results demonstrate that our proposed method can get very competitive results while maintaining real-time performance.