Research on ARC Discharge Detection in Switchgear Based on Improved YOLOv11

Arc discharge faults in switchgear cabinets present critical challenges to power systems due to their stochastic occurrence, concealed characteristics, and complex physical mechanisms. To address the limitations of conventional detection methods in feature ambiguity and response latency, this paper proposes an enhanced YOLOv11-based intelligent detection framework. Three core innovations are implemented: 1) A deformable convolutional network (DCNv4) is integrated into the backbone, dynamically adapting receptive fields to capture irregular discharge morphologies; 2) A hybrid CNN-Transformer architecture synergistically fuses local spatial features with global contextual dependencies; 3) An adaptive bounding box regression strategy with parametric IoU loss optimization enhances multi-scale detection robustness. Experimental validation demonstrates superior performance, achieving 3.94% and 3.67% improvements over YOLOv8 and YOLOv10 respectively, confirming practical applicability in complex grid scenarios.