Detecting Road Conditions from Space Using PyTorch, Public Data, and Free Satellite Imagery

Monitoring the health of city streets and roadways is expensive, time-consuming, and often reactive. But what if we could automate part of that process using satellite imagery and Python?

In this talk, we’ll walk through a real-world project that combines transfer learning, PyTorch, and publicly available datasets to classify road segment conditions (good, fair, poor) from aerial imagery. You'll learn how to work with messy real-world geospatial data, fine-tune a deep learning model using only a small training set (~2,000 examples), and overcome common challenges like blurry imagery, inconsistent labels, and overfitting.

This session is practical and code-driven, aimed at data scientists and analysts working in mobility analytics, urban development, or infrastructure who want to apply machine perception techniques in their work. By the end, you’ll walk away with a reusable workflow for analyzing and predicting urban infrastructure quality — all using free tools and open data.