Machine Learning (ML) can be used to supercharge human mapping efforts by prioritizing which areas to look in. Instead of generating geometries, ML models can estimate the likelihood that a feature appears in a certain area. Then human mappers can use this guidance to decide where they map - focusing on quality, and improving efficiency.
We'll walk through building a machine learning pipeline and present a workflow for human mappers to create OpenStreetMap features over large areas. The workshop will cover:
- Using OpenStreetMap as training data for a ML model (using the open source tool Label Maker (https://github.com/developmentseed/label-maker))
- Training a sophisticated ML model, to identify whether certain features appear in a tile
- Using the output of the model to help guide mappers to priority areas
- Integrating this workflow into existing OpenStreeMap tools (Tasking Manager and To-Fix)
We've used this method to map electricity infrastructure in Pakistan, Nigeria, and Zambia. We look forward to sharing it with more mappers and increasing the impact of everyone's efforts.