An Electric Grid Sign Detection Method Based on the Improved YOLOv8n

To tackle the common issue of signboard detection in power grid equipment inspections, this paper introduces an enhanced method for power grid signboard detection based on YOLOv8n. By integrating the Global Attention Mechanism (GAM) into the YOLOv8n network architecture, the deep neural network's performance is boosted, mitigating information loss and enhancing global feature interaction. Moreover, the Focal-SIOU loss function replaces CIOU for bounding box regression, leading to improved convergence speed, accuracy, robustness, and adaptability across various scenarios and conditions. The proposed YOLOv8n algorithm is evaluated on a custommade power grid signboard dataset. The results demonstrate that the enhanced YOLOv8n achieves an average precision mAP0.5 of 95.4% on the power grid signboard dataset, marking a 1.6% enhancement compared to YOLOv8n. This advancement significantly improves the accuracy of power grid signboard detection.