Dual Graph Reasoning Network for Oil Leakage Segmentation in Substation Equipment

Oil leakage inspection is critical to ensure the safety of substation operations. However, weak edges and irregular segmentation pose challenges in oil leakage inspection tasks. To address these challenges, this article proposes a semantic segmentation network based on a dual graph reasoning mechanism, named as DGRNet. First, the spatial graph reasoning (SpGR) branch learns edge detail information by modeling long-term dependencies between pixels and employs auxiliary losses to further improve edge awareness. Second, the semantic graph reasoning (SeGR) branch learns the global context by modeling the global relationships between regions and reinforces texture feature representations. Finally, the dual branch structure is adopted to optimize the fusion mechanism of deep features and shallow features, achieving accurate segmentation of the oil leakage area. The experimental results demonstrate that the proposed network achieves significant performance improvement on the custom substation equipment oil leakage segmentation dataset, surpassing classical semantic segmentation networks. The intersection over union (IOU) and the
$F1$
score (
$F1$
) of oil leakage reach 73.77% and 84.91%, respectively, with the pixel accuracy (PA) reaching 92.19%.