Large-scale School Mapping using Weakly Supervised Deep Learning for Universal School Connectivity

TL;DR

The paper tackles the critical need for complete school geolocation data to plan universal connectivity. It introduces a scalable weakly supervised deep learning pipeline that combines vision transformers and CNNs, using class activation maps to localize schools from image-level labels in high-resolution satellite imagery. The approach achieves strong performance (AUPRC > 0.96) across 10 African countries, enables nationwide predictions, and includes an interactive validation tool for government partners, underscoring practical applicability. By leveraging diverse data sources and explainable AI, the method supports cost-effective, near-real-time planning of school connectivity and highlights the importance of local data and stakeholder collaboration for successful deployment.

Abstract

Improving global school connectivity is critical for ensuring inclusive and equitable quality education. To reliably estimate the cost of connecting schools, governments and connectivity providers require complete and accurate school location data - a resource that is often scarce in many low- and middle-income countries. To address this challenge, we propose a cost-effective, scalable approach to locating schools in high-resolution satellite images using weakly supervised deep learning techniques. Our best models, which combine vision transformers and convolutional neural networks, achieve AUPRC values above 0.96 across 10 pilot African countries. Leveraging explainable AI techniques, our approach can approximate the precise geographical coordinates of the school locations using only low-cost, classification-level annotations. To demonstrate the scalability of our method, we generate nationwide maps of school location predictions in African countries and present a detailed analysis of our results, using Senegal as our case study. Finally, we demonstrate the immediate usability of our work by introducing an interactive web mapping tool to streamline human-in-the-loop model validation efforts by government partners. This work successfully showcases the real-world utility of deep learning and satellite images for planning regional infrastructure and accelerating universal school connectivity.

Figures (7)

• Figure 1: (Left) 300 x 300 m (500 x 500 px) satellite image tile of a school in Senegal. (Middle) GradCAM outputs. (Right) Perturbation of the top 10% of pixels using ROAD.
• Figure 2: Per-country AUPRC by urban/rural subgroup.
• Figure 3: A comparison of the AUPRC of the best regional model versus the best local model per country.
• Figure 4: Pairwise train/test AUPRC.
• Figure 5: A comparison of model predictions and government data for Senegal ($\tau=0.5$ and $d=250 m$).