GridTracer: Automatic Mapping of Power Grids Using Deep Learning and Overhead Imagery

Energy system information for electricity access planning such as the locations and connectivity of transmission distribution towers—termed power grid—is often incomplete, outdated, or altogether unavailable. Furthermore, conventional means collecting this is costly limited. We propose to automatically map grid in overhead remotely sensed imagery using an deep learning approach. Toward goal, we develop publicly release a large dataset (263 km$^2$) with ground-truth grid—to our knowledge, first its kind public domain. Additionally, scoring metrics baseline algorithms two grid-mapping tasks: 1) tower recognition 2) line interconnection (i.e., estimating graph representation grid). hope availability training data, metrics, baselines will facilitate rapid progress on important problem help decision-makers address energy needs societies around world.