Deep learning-based detection of power transmission lines using YOLOv4 and YOLOv8

Area of Science:

• Electrical Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Power transmission line (PTL) maintenance is crucial for reliable energy distribution.
• Traditional PTL inspection methods are inefficient, costly, and pose safety risks.
• Unmanned aerial vehicle (UAV) technology offers enhanced efficiency, safety, and cost-effectiveness for infrastructure monitoring.

Purpose of the Study:

• To evaluate the effectiveness of deep learning object detection, specifically the You Only Look Once (YOLO) algorithm, for PTL inspection using UAVs.
• To compare the performance of different YOLO versions in detecting power transmission lines.

Main Methods:

• Utilized UAVs equipped with cameras for aerial data acquisition.
• Applied various versions of the YOLO algorithm for object detection and classification of PTLs.
• Analyzed performance metrics including precision, recall, F1 score, mAP50, and mAP50-95.

Main Results:

• All tested YOLOv8 versions demonstrated superior performance over YOLOv4 for PTL detection.
• YOLOv8 achieved precision, recall, and F1 scores exceeding 99%.
• Average mAP50 and mAP50-95 values were recorded at 0.995 and 0.919, respectively.

Conclusions:

• Deep learning-based object detection with YOLOv8 integrated with UAVs provides a highly accurate and efficient solution for PTL monitoring.
• YOLOv8 significantly outperforms previous versions, establishing a new benchmark for PTL inspection technology.
• This approach enhances the safety and reliability of energy distribution networks.