Improved YOLOv8 Algorithm for Detecting Small Target Fittings on Transmission Lines

Targeting the problem of a small number of pixels of small targets and lack of effective feature information, difficult background differentiation, and sensitivity to localization offset in the task of transmission line small target fittings detection, this paper proposes an algorithm for transmission line small target fittings detection, which is noted as SCCN-YOLO. Based on YOLOv8, this paper firstly adds a small target detection layer, which enhances the fusion effect of the small target features; secondly, it uses the CARAFE upsampling operator in the backbone network to expand the sensory field of the upsampled features, improving the network's ability to extract small target features, capturing semantic information, and reducing information loss are enhanced by this approach; Then the CBAM attention mechanism is introduced to focus on the target features to inhibit the interference of irrelevant information and to enhance the small target features; Finally, the NWD loss function is used to reduce the sensitivity of the loss function to the target localization bias and to improve the network's ability to detect. The experimental results prove that the accuracy of the algorithm proposed in this paper reaches 87.9%, and it can effectively detect the small target fittings of transmission Lines.