Christopher is a Class of 2015 graduate of Maui High School and will be a junior at Yale University in the Fall of 2017. He is working towards B.S. degrees in computer science and economics and is particularly passionate about the potential of computer science in opening new frontiers. He hopes to utilize his higher education to pioneer advancements that will enrich the lives of others. At Yale, Chris is involved with Yale’s student-run hackathon, YHack, teaches computer science to New Haven middle school students, is a teaching assistant for Yale’s introductory computer science courses, and is an active member of an on-campus Christian fellowship.
High School: Maui High School
Institute when accepted: Yale University
Localization of Closely-Spaced Deep Space Objects Using High Frame Rate Imagery via Deep Learning
Project Site: Air Force Maui Optical and Supercomputing Site
Mentors: Capt. Justin Fletcher, 1st Lt. Julian McCafferty
Space situational awareness is the discipline associated with understanding the state of the space environment and objects in that environment. Deep space closely-spaced object localization (C-SOL) is one problem within this domain. C-SOL is the problem of determining the relative locations of two resident space objects in geosynchronous orbit with dissimilar brightness in close proximity to one another. Previous investigations demonstrated that separation information may be extracted from a sequence of high frame rate images. These efforts did not yield software modules capable of localizing CSOs. High frame rate imagery captures the instantaneous state of atmospheric turbulence and contains information about the location of objects that would be lost due to the accumulation of turbulence-derived signal perturbations. In this work, we demonstrate the applicability of machine learning techniques to the C-SOL problem. As real data from this domain is fairly rare, simulation of the relevant phenomena is required. Thus, this work required the transliteration and modernization of legacy MATLAB code that performed physics-based simulations of high frame rate imagery into Python. The produced software generates and organizes high frame rate imagery data into datasets to be used for training and empirical validation of machine learning models. These datasets are used train a convolutional neural network to extract salient features and localized objects using those features. The findings of this work suggest that deep learning is a viable means of solving the C-SOL problem.