Aerial identification of flashed over faulty insulator using binary image classification

Abstract

Flashed over insulator faults are the most significant faults in high voltage line insulators. They are complicated to identify using traditional methods due to their labor-intensive nature. This study proposes a deep learning-based algorithm for detecting flashed over insulator faults in the real time. The algorithm is based on the Resnet 50 architecture, which has been shown to be effective for image classification tasks in the previous studies regarding image analysis. The algorithm is fast, robust and efficient, making it suitable for real-time applications. The algorithm is trained on a dataset of images of flashed over and non-flashed over insulators. This dataset was collected from various transmission lines and National Center of Robotics and Automation, which are located in Pakistan. For validating the effectiveness of the Resnet 50 algorithm, it was compared with the results obtained from the two other widely popular deep learning algorithms, Densenet 121 and VGG 16 (trained and validated on the same dataset). The results showed that the Resnet 50 was able to detect flashed over insulator faults with an accuracy of over 99%. Whereas the Densenet 121 and VGG 16 have achieved an accuracy of less than 51%.