Galactic Flood Fill Segmentation and Machine Learning Redshift Estimation
| dc.contributor.author | Ferguson, Matthew Chase | en |
| dc.contributor.committeechair | Jones, Creed Farris | en |
| dc.contributor.committeemember | Talty, Timothy Joseph | en |
| dc.contributor.committeemember | Plassmann, Paul E. | en |
| dc.contributor.department | Electrical and Computer Engineering | en |
| dc.date.accessioned | 2025-01-22T09:00:15Z | en |
| dc.date.available | 2025-01-22T09:00:15Z | en |
| dc.date.issued | 2025-01-21 | en |
| dc.description.abstract | This thesis explores the use of machine learning redshift estimation models trained on segmented galactic images. Segmentation of galaxies from the background is accomplished using a flood fill segmentation method which is novel to the field of galactic segmentation. Astronomy datasets are so large due to high volume modern surveys that automated analysis techniques are now required. Redshift is a prime example of an expensive measurement that is a candidate for automation. The Sloan Digital Sky Survey alone imaged more than 1 billion objects in 9 years, but only produced 4 million spectra over more than 20 years. Machine learning is an automation technology that promises to efficiently analyze imaging data alone such that redshift can be estimated with a high degree of accuracy. Ground truth redshift and multi-band galactic images were obtained for 200,000 galaxies from the Sloan Digital Sky Survey. Two model architectures were experimented with, a fully connected artificial neural network, and a convolutional neural network. Experiments were conducted on flood fill parameters, crop sizes, color spaces, and thresholding. We demonstrated that model performance on flood fill segments is higher than on unsegmented images across many crop sizes. The best achieved model performances for artificial neural networks, and convolutional neural networks are median absolute dispersions of 0.024 and 0.031, respectively. | en |
| dc.description.abstractgeneral | Astronomy has progressed to the point where there is too much data for humans to analyze manually. Every year more data is created. Machine learning is a technology that can sort through all this data. We used machine learning to create a method to predict redshift of galaxies. Redshift is a measure of the object's speed and tells us about the history of the universe. We also created a novel way of separating galaxies in images from the background using flood fill. Machine learning creates a function that connects two datasets. Our first dataset is 200,000 galactic images from the Sloan Digital Sky Survey, and our second dataset is redshift measurements of these galaxies. We connected images to their redshift using a machine learning model. This lets us take images of galaxies and estimate their redshift. Our best models exceeded the performance of some other redshift estimation techniques. | en |
| dc.description.degree | Master of Science | en |
| dc.format.medium | ETD | en |
| dc.identifier.other | vt_gsexam:42354 | en |
| dc.identifier.uri | https://hdl.handle.net/10919/124287 | en |
| dc.language.iso | en | en |
| dc.publisher | Virginia Tech | en |
| dc.rights | In Copyright | en |
| dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | en |
| dc.subject | Redshift | en |
| dc.subject | Flood Fill | en |
| dc.subject | Segmentation | en |
| dc.subject | Machine Learning | en |
| dc.subject | Estimation | en |
| dc.subject | Regression | en |
| dc.subject | Galaxy | en |
| dc.subject | Deep Learning | en |
| dc.title | Galactic Flood Fill Segmentation and Machine Learning Redshift Estimation | en |
| dc.type | Thesis | en |
| thesis.degree.discipline | Computer Engineering | en |
| thesis.degree.grantor | Virginia Polytechnic Institute and State University | en |
| thesis.degree.level | masters | en |
| thesis.degree.name | Master of Science | en |
Files
Original bundle
1 - 1 of 1