PTL-AI Furnas Dataset: A Public Dataset for Fault Detection in Power Transmission Lines Using Aerial Images
Frederico Santos de Oliveira, Marcelo de Carvalho, Pedro Henrique Tancredo Campos, Anderson da Silva Soares, Arnaldo Cândido, Ana Claudia Rodrigues Da Silva Quirino
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.