Improving OpenStreetMap missing building detection using few‐shot transfer learning in sub‐Saharan Africa

Abstract OpenStreetMap (OSM) has been intensively used to support humanitarian aid activities, especially in the Global South. Its data availability South greatly improved via recent mapping campaigns. However, large rural areas are still incompletely mapped. The timely provision of map is often essential for work actors case disaster preparation or response. Therefore, it become a vital challenge boost speed and efficiency existing workflows. We address this by proposing novel few‐shot transfer learning (FSTL) method improve accuracy OSM missing building detection. trained two popular object detection models (i.e., Faster R‐CNN SSD) training area Tanzania transferred model target Cameroon Mozambique. FSTL significantly base performance even with only one shot. Moreover, we successfully produced grid‐based (DeepVGI) 10 m spatial resolution over 96% overall (ACC) 0.85 Matthews correlation coefficient (MCC) both Such maps show great potential assess estimate completeness buildings places where other (e.g., buildings, roads) datasets not available.