Bob Bell, Researcher with the Urban Observatory Project at the SFS, led both lectures that familiarized attendees with QGIS software — a free, open-source desktop geographic information system application that lets users view, edit, and analyze geospatial data across platforms. Rajesh Veeraraghavan, Assistant Professor of Science, Technology, and International Affairs (STIA), described how satellite imagery data can be used to understand urban settlement characteristics, including access to services.
The first workshop, “Mapping Infrastructure Categories from GIS Satellite Imagery” was held on April 15 and was tailored to beginners that had no prior experience with QGIS. It provided practical tools and instructions for mapping informal settlements with satellite imagery using QGIS after users downloaded the system application. By drawing on his own research on Indian government-designed categories of unplanned settlements, Bell introduced attendees to vector and raster layers, helped them create and import shapefiles, and provided methods to access and identify settlements using remote sensing satellite imagery.
The second workshop in this series, held on April 16, focused on classifying infrastructures within satellite imagery with machine learning. This workshop provided a practical introduction to machine learning for computer vision on remote sensing imagery using Python and the PyTorch deep learning framework. It introduced attendees to remote sensing imagery used in Bell’s own research, which includes data sources and different forms of resolution (e.g., spatial, temporal, radiometric, spectral). The session demonstrated how to preprocess imagery–including rasterizing polygons, tiling rasters into image chips, and image resizing and augmentation–before transitioning to machine learning. It defined training and validation sets, classified imagery with a neural network in PyTorch, and assessed the accuracy of the classifier on a validation set. Attendees used Python to run the machine learning algorithms, building a Jupyter notebook of the entire learning task.
Recent developments in machine learning provide researchers with unprecedented opportunities to harness remote sensing for sustainable development and humanitarian efforts. And with developing countries increasingly urbanizing, it is critical to use satellite imagery software to understand and track the evolution and spatial characteristics of informal settlements for both poverty targeting efforts and the provision of public services.