Detecting Rural Habitat using Satellite Imagery

Uganda


Client Brief

  • Knowing where people live and counting them is crucial in many of our assignments, more specifically in rural areas (to detect underserved populations for instance)

  • Some datasets that granularly estimate the population are already available but are often incomplete or imperfect


Solution

  • Masae Analytics used a Convolutional Neural Network trained with roughly 14,000 images distributed between presence/absence of house


Results

  • The convolutional neural network yielded a 97% accuracy score in detecting human settlements

  • Results, especially in rural areas, were significantly more accurate than many other alternatives