Electric Pylon Detection in Various Environmental Conditions Through High-Resolution Satellite Images

Earth remote sensing data can be applied to detect and assess the condition of infrastructure objects on vast territories. One such object is electric pylons, which ensure the sustainability of the energy supply in rural and urban areas. In some remote regions, power lines can be damaged by natural hazards such as earthquakes, strong winds, or floods. Currently, the main limitation in developing highly effective algorithms for electric utilities assessment is associated with data availability and diverse environmental conditions. Therefore, in this study, we aim to explore solutions for new study territories with various backgrounds and forms of electric pylons. We examined several detection algorithms from the YOLO-family. The study includes experiments with datasets for Chinese territories and additionally collected data for regions in Russia. We managed to improve the initial score for polygon detection, achieving an mAP of 79.8%. The obtained results demonstrate high potential for power lines assessment and damage detection through satellite data and deep learning algorithms.