To implement effective rural electricity access systems, it is fundamental to identify where potential consumers live. Here, we test the suitability of citizen science paired with satellite imagery and computer vision to map remote off-grid homes for electrical system design. A citizen science project called “Power to the People” was completed on the Zooniverse platform to collect home annotations in Uganda, Kenya, and Sierra Leone. Thousands of citizen scientists created a novel dataset of 578,010 home annotations with an average mapping speed of 7 km2/day. These data were post-processed with clustering to determine high-consensus home annotations. The raw annotations achieved a recall of 93% and precision of 49%; clustering the annotations increased precision to 69%. These were used to train a Faster R-CNN object detection model, producing detections useful as a first pass for home-level mapping with a feasible mapping rate of 42,938 km2/day. Detections achieved a precision of 67% and recall of 36%. This research shows citizen science and computer vision to be a promising pipeline for accelerated rural home-level mapping to enable energy system design.
