Identifying clouds with neural networks

A collaborative research project between the computer science and environmental studies department at Lewis & Clark College. The goal of the research is to automate the identification of of meteorological clouds given a photographic image. Myself and a peer wrote the code to implement and train a convolutional neural network in addition to managing and preprocessing the image data. The network was tasked with distinguishing between sky, thin cloud pixels, and thick cloud pixels; the trained network achieved a 91.9% classification accuracy and an ensemble achieved 94.6% accuracy. This research was in collaboration with faculty advisors Dr. Peter Drake and Dr. Jessica Kleiss of Lewis & Clark College. Thanks to the John S. Rogers Science Program for funding this research.