ML-assisted Defect Detection in Power line Insulators for Reducing Energy Wastage

Power line insulator defects pose significant risks to the reliability and safety of electrical distribution systems. To provide continuous power supply and preventive maintenance, these problems must be promptly detected. It is common practice to employ an airborne platform-based insulator inspection technique to find the issues. Detecting insulator defects in aerial photographs against complicated backdrops is intriguing yet difficult. This work proposes a novel method for automating the detection of Power line insulator flows using the You Only Look Once version 5 (YOLOv5) object identification algorithm. The proposed method includes Transfer learning, data preparation approaches, and augmentation tactics, which shows the promising potential to improve the insulator fault detection model’s accuracy and recall rates. The proposed method also uses post-processing methods to improve problem localization further and eliminate false positives. We assess the effectiveness of the suggested approach using a real-world dataset made up of pictures gathered from several electricity distribution networks. The outcomes of the experiments demonstrate the potential of the proposed method to precisely and recall rates detect insulator faults.