PTL-AI Furnas Dataset: A Public Dataset for Fault Detection in Power Transmission Lines Using Aerial Images

Abstract

We present a new images dataset called PTL-AI Furnas Dataset as a new benchmark for fault detection in power transmission lines. This dataset has 6,295 images, with resolution $1280\times 720$, extracted from the maintenance process of the energy transmission lines at Furnas company. It contains annotations of 17,808 components classified as baliser, bird nest, insulator, spacer and stockbridge. Furnas is a company that generates or transmits electricity to 51% of households in Brazil and more than 40% of the nation's electricity passes through their grid enabling generating the dataset in different backgrounds and climatic conditions. We performed experiments using data augmentation techniques to train Faster R-CNN, Single-Shot Detects (SSD) and YoloV5 models. The benchmark result was obtained using the metrics of Mean Average Precision (mAP) and the Mean Average Recall (mAR) with values mAP=91.9% and mAR=89.7%. The PTL-AI Furnas Dataset is publicly available at https://github.com/freds0/PTL-AI_Furnas_Dataset.