YOLO-SS-Large: A Lightweight and High-Performance Model for Defect Detection in Substations

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Computer Vision

Background:

• Active distribution grids are increasingly adopting renewable energy sources.
• Substations in these grids face challenges with timely defect detection and slow inspection responses.
• These issues pose risks to the safe and stable operation of active distribution grids.

Purpose of the Study:

• To propose a high-performance, lightweight model for substation defect detection.
• To enhance the detection of small object defects in complex substation environments.
• To improve the overall efficiency and accuracy of intelligent substation inspections.

Main Methods:

• Development of the YOLO-SS-large model based on YOLOv5m, incorporating the FasterNet structure for improved lightweight performance.
• Introduction of a novel NWD-CIoU loss function, fusing Normalized Wasserstein Distance (NWD) and Complete Intersection over Union (CIoU) for enhanced small object detection.
• Integration of a dynamic head module to unify scale-aware, spatial-aware, and task-aware attention mechanisms.

Main Results:

• The YOLO-SS-large model achieved a 0.3% average precision improvement compared to YOLOv5m.
• Demonstrated a 43.5% enhancement in Frames Per Second (FPS), indicating increased speed.
• Achieved a 41.0% reduction in parameters, highlighting its lightweight nature.

Conclusions:

• The YOLO-SS-large model offers significantly enhanced comprehensive performance for substation defect detection.
• The model effectively meets the speed and precision requirements for intelligent substation inspections.
• This advancement is vital for promoting the informatization and intelligent construction of active distribution grids.