SONet: A Small Object Detection Network for Power Line Inspection Based on YOLOv8
Weicheng Shi, Xiaoqin Lyu, Lei Han
Abstract
Power line inspection plays a crucial role in ensuring the security of power systems, and the difficulty in detecting small objects is one of the main problems in power line inspection. This paper proposes a small object detection network for power line inspection based on YOLOv8, which is called SONet. Firstly, a multi-branch dilated convolution module (MDCM) is proposed, which can obtain multiple features in different receptive fields and thus enrich the features of small objects. Secondly, an adaptive attention feature fusion structure (AAFF) is proposed to replace the PANet, which can guide the feature fusion by adaptive attention and improve the effect of the feature fusion while reducing the number of parameters. Thirdly, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">β</i>-CIoU loss is proposed to dynamically optimize the learning rate during bounding box regression, thereby enhancing the detection accuracy of small objects. The results indicate that the proposed model's detection accuracy reaches 78.67%, and the small object detection accuracy reaches 20.0%. The detection speed reaches 32.5 FPS. The results verify the effectiveness of the proposed method in the task of small object detection for power line inspection.