Lightweight SCL-YOLOv8: A High-Performance Model for Transmission Line Foreign Object Detection.

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Transmission lines face risks from foreign object intrusion, potentially causing power outages.
• Current UAV-based inspections lack real-time analysis and efficiency due to data transmission needs.
• Existing YOLO models are too computationally intensive for efficient UAV deployment.

Purpose of the Study:

• To develop a lightweight foreign object detection algorithm for real-time analysis on UAV platforms.
• To improve the efficiency and accuracy of foreign object detection on transmission lines.
• To reduce the computational complexity and model size of object detection systems for UAVs.

Main Methods:

• Proposed SCL-YOLOv8, a novel lightweight detection algorithm based on YOLO.
• Replaced the CSPDarknet53 backbone with StarNet for reduced computational complexity and efficient feature extraction.
• Introduced a CGLU-ConvFormer module for enhanced multi-scale and local feature extraction.
• Improved the detection head with shared convolutional layers and group normalization for reduced computation and better feature fusion.

Main Results:

• SCL-YOLOv8 achieved a mean Average Precision (mAP@0.5) of 94.2%.
• Significantly reduced model parameters by 56.8%, Floating Point Operations (FLOPs) by 45.7%, and model size by 50% compared to YOLOv8n.
• Demonstrated improved detection accuracy and reduced computational requirements.

Conclusions:

• SCL-YOLOv8 offers an efficient and accurate solution for real-time foreign object detection on transmission lines via UAVs.
• The lightweight design enables deployment on resource-constrained UAV platforms.
• This advancement contributes to enhanced power grid safety and operational efficiency.