SDS-YOLO: drone-based foreign object detection model for power lines using an enhanced YOLOv8n approach.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• UAV-based transmission line inspection faces challenges with foreign object detection accuracy and model efficiency.
• Existing methods struggle with lightweighting models while maintaining high detection performance.

Purpose of the Study:

• To develop an optimized foreign object detection algorithm (SDS-YOLOv8n) for UAV-based transmission line inspection.
• To enhance detection accuracy and model lightweighting for real-time intelligent grid inspection.

Main Methods:

• The proposed SDS-YOLOv8n is built upon the YOLOv8n architecture with three key improvements.
• Incorporates an Enhanced SPPF module for better focus on target details and reduced background interference.
• Replaces the standard detection head with a Dynamic Detection Head featuring unified attention mechanisms and integrates a Parameter-free Attention Mechanism (SimAM).

Main Results:

• SDS-YOLOv8n achieved a mAP@0.5 of 95.8% and mAP@0.5:0.95 of 75.1%, outperforming the baseline.
• Demonstrated strong generalization capabilities on an untrained dataset.
• The model has a parameter count of 2.75 M and high inference speed on the NVIDIA Jetson Orin Nano platform.

Conclusions:

• SDS-YOLOv8n offers a robust and efficient solution for real-time intelligent grid inspection.
• The proposed optimizations effectively balance detection accuracy and computational efficiency.
• The algorithm provides a significant advancement for UAV-based transmission line monitoring.