The School of Computer Science is pleased to present…

Classifying Galaxy Morphologies: Improved Residual Networks 

MSc Thesis Proposal by: Jaykumar Patel 

Date: Thursday, December 15, 2022 

Time: 11:30 am-1:00 pm 

Location: Essex Hall, Room 122

Reminders: Two-part attendance is mandatory; arrive 5-10 minutes prior to the event starting – LATECOMERS WILL NOT BE ADMITTED once the door has been closed and the presentation has begun. Please be respectful of the presenter by NOT knocking on the door for admittance.

Abstract:  

For the classification of images of galaxy morphology (a term used by astronomers to categorize galaxies based on how they appear visually), a variant of Residual Neural Network (RNN) was developed in this research. The variant, together with other popular convolutional neural networks (CNNs), is applied to a sample of 28790 and 25941 galaxy images from the Galaxy Zoo 2 dataset to classify galaxies into five and seven classes, respectively. The five classes are categorized as Completely round, In-between, Cigar shaped, Edge-on, and Spiral, and the seven classes are categorized as Completely round, In-between, Cigar shaped, Edge-on, Barred spiral, Unbarred spiral, and Irregular. Various metrics, such as accuracy, precision, recall, and F1 score, show that the proposed network achieves state-of-the-art classification performance among other networks, namely, VGG16, VGG19, Inception, ResNets, and EfficientNets. The overall classification accuracy of our network on the testing sets is 96% and 93% for datasets with five and seven classes, respectively. The variant of the Residual Network model proposed in this study can be applied to large-scale galaxy classification in optical space surveys, which provides a large amount of data.

Keywords: Morphology, Neural Networks, Galaxy Zoo

MSc Thesis Committee:  

Internal Reader: Dr. Saeed Samet

External Reader: Dr. Dan Xiao

Advisor: Dr. Dan Wu