Collaborative Detection of Power Equipment Failures and Worker Protection for Hydropower Stations: Research on the Lightweight DRR-YOLOv11s Model

The growing demand for intelligent operation and maintenance of hydropower plants highlights the limitations of traditional detection methods. These methods cannot fully satisfy the demand for real-time capability, efficiency, and accuracy in equipment fault identification and ensuring worker safety. To resolve this problem, we present DDR-YOLOv11s, a lightweight detection algorithm based on YOLOv11s. The model integrates three efficient modules: DDSConv, which replaces standard convolutions to reduce the amount of parameters and FLOPs; RDSC, an improved detection head optimized for small-object detection; and enhanced RepViTBlocks, which substitute C3K2/C3K bottlenecks to strengthen feature representation and information exchange. DDR-YOLOv11s demonstrates significant lightweight performance. Relative to the baseline model, the parameter count decreased by 37.58% and the computational cost was lowered by 28.64%, yielding a total of 5.88M parameters and 15.2G FLOPs. On the Electrical Equipment Failure dataset (Roboflow), it achieves a precision of 92.50%, a recall of 89.42%, an mAP@0.5 of 93.46%, and an mAP@0.5:0.95 of 73.45%, with an inference speed of 160.22 FPS. Cross-dataset evaluation on a PPE dataset confirms its strong generalization ability. These results indicate that DDR-YOLOv11s attains a favorable balance between accuracy and computational cost, highlighting its applicability for edge deployment in hydropower monitoring systems.