Hugh Shields ’24
Civil and Environmental Engineering
Extinction of the Dinosaurs Recorded in an Andean Paleolake
I primarily focused on ooids, grains of calcium carbonate that form in shallow waters. Ooids grow by precipitation and shrink by abrasion. For this reason, their shape can reveal data about ancient ocean chemistry, currents and depth. I analyzed cross sections of rock samples containing ooids, specifically trying to develop a machine vision program that would outline ooid cross sections within the images. Individual cross sections could then be stacked to produce a 3D model of the entire ooid, allowing for the measurement of sphericity and ellipticity. To complete the segmentation task, I used the software TensorFlow to implement Mask Region Based Convolutional Neural Networks (Mask R-CNN), a machine learning model for object detection. After tuning the network, I finally reached the point where the model was obtaining relatively high accuracy scores, and I began to transition toward applying the network to a wider variety of samples. The model I trained revealed that segmentation could be done successfully using Mask R-CNN without hours of laborious manual tracing. Ultimately, the project gave me the theoretical and practical experience needed to implement a cutting-edge machine vision model and revealed the wide applicability of machine vision.
Climate and Environmental Science
Maloof Research Group, Department of Geosciences, Princeton University
Adam Maloof, Professor of Geosciences; Bolton Howes, Ph.D. candidate, Geosciences