Image Recognition Applied to Insulator Detection and Classification for Asset Management

Overhead transmission lines are critical to power infrastructure, with insulators playing a pivotal role in ensuring network reliability by preventing electrical leakage or electricity conductors falling to the ground. Traditional methods of insulator assessment rely heavily on manual labour-intensive and time-consuming audits, which could be more efficient given the vast scale of modern transmission networks. This paper presents an innovative solution leveraging machine learning (ML) and image recognition to automate insulator detection, identification, and location within aerial images of transmission towers. By retraining convolutional neural networks (CNN), specifically YOLO, we address the challenges posed by diverse image conditions, unbalanced datasets, and the specificity of insulator attributes such as material, type, and arrangement. Our original approach, which combines multiple models in sequence and in parallel, improves detection accuracy and streamlines the management of electricity transmission assets by reducing manual intervention. Training and combining these ML models allows us to reach an f1-score of around 97.5%. We integrate our approach with existing asset management systems, providing real-time data for more informed decision-making. The results show significant potential for enhancing transmission line maintenance and reliability through automated insulator management.