Arc Image Dataset for Power Grid Fault Detection (PG-ArcSet)

With the expanding adoption of deep learning in power system monitoring, automated fault diagnosis based on arc image analysis has attracted significant research interest. However, the absence of standardized, well-annotated arc image datasets has hindered the development and fair evaluation of related recognition algorithms. To address this gap, we introduce a high-voltage arc image dataset comprising three fundamental arc types: line-type, branch-type, and net-type. The arcs are simulated under controlled laboratory conditions, consistent with physical discharge mechanisms. The dataset includes 1,950 preprocessed images, where each arc instance is centered and resized to 224$\\times$224 pixels, making it suitable for training deep learning models. Additionally, we provide the Python-based preprocessing scripts used to generate the final dataset. These scripts can be adapted to promote arc data standardization and facilitate reproducible research in power equipment fault diagnosis.