CWA-YOLO: Rethinking the YOLOv8 Architecture at the Level of Feature Information and Applying it to Substation Equipment Detection

Substations act as bridges connecting power gener-ation with consumption and play crucial roles in electrical grids such as power energy conversion, operational dispatching, and information monitoring. With the rapid development of electrical grids in recent years, the increasing contradiction between the shortage of production staff and the growing workload of inspections has urgently necessitated intelligent recognition algorithms for the condition of indoor substation equipment. This paper proposes a novel indoor substation equipment target detection framework based on YOLOv8, aimed at addressing issues like scale variation, occlusion handling, and real-time performance optimization, thereby enhancing detection accuracy and speed in complex environments. By incorporating the ConvNeXt feature extraction network and BiFPN, combined with the Wise-IoU loss function and adding detection heads for small targets, this approach has improved the detection capabilities for multi-scale targets, particularly in recognizing minute objects. Experimental results with actual field data indicate that the proposed CWA-YOLO model outperforms existing methods in both detection accuracy and speed, with the model mAP0.5 and mAP0.5:0.95 being 0.918 and 0.623 respectively, marking improvements of 6.50% and 11.05% over the YOLOv81 model, 7.49% and 11.85% over YOLOv71, and 9.55% and 15.80% over YOLOv51.