Astrophysics Undergraduate Thesis

A repository containing jupyter notebook interface of a Convolution Neural Network Model to identify stellar wind bubbles in galaxies.

This model is trained on galaxy NGC 0628 as it is the only galaxy who has been subjected to a stellar-wind bubble analysis ( J. Watkins et al 2023 ApJL )

Basic CNN model

To get a taste of how convolutional neural network works, directory CNN training provides a good overview of the setup.

1. jwst-bubbles-v0p1.reg contains information regarding the bubbles
2. ngc0628_miri_f770w_anchored.fits is a fits file which contains the image of the galaxy NCG 0628 and can be easily plotted using astropy library in python.

U-NET architecture

U-Net contains 4 directories:

1. 0-1 dir showcases a technique where overlapping bubble regions are treated as a single bubble large bubble with a not elliptical or a circular morphology

   1.1 Main emphasis of the research study was improving this part of the spectrum

   1.2 Subdir 'Preparing DataSets' consists of python automation scripts capable of preparing training images.

2. '>1' takes into account overlapping region (needed a completely new research study as U-NET model based on 0-1 technique didn't perform as well)

3. NGC0628 dir contains the ground truths to train the model

4. dataset dir provides an example of the training set

Every bit of this code is adaptable according to the idiosyncratic nature of study.