A team of students led by researchers in the Energy Data Analytics Lab and the Sustainable Energy Transitions Initiative will develop a method to evaluate electricity access in developing countries through machine learning techniques applied to aerial imagery data. Students will first improve the accuracy of the solar array identifying U-Net model, a convolutional neural network for image segmentation, built by students in the Fall 2024/Spring 2025 Bass Connections course. In addition, students will develop a new model that identifies solar water heaters. These two models can be used to demonstrate whether a community has access to electricity, create a reference dataset of key features, and apply machine learning methods to a large dataset. This work will provide a needed basis for research groups at Duke and elsewhere interested in understanding the path to electrification in under served areas of Cape Town and may result in more comprehensive maps of electricity access and assets.
Project Lead: Marc Jeuland (Public Policy)
Project manager; Biz Yoder (Public Policy)
