A Neural Network Approach Using Deep Learning for Image Classification of Polar Ring Galaxies

Overview

Understanding the formation of peculiar galaxies, such as polar ring galaxies, can aid in learning the properties of dark matter and the process of deciphering how galaxies evolved. A polar ring galaxy is a type of galaxy with a ring of gas and star matter orbiting over the perpendicular plane of its host galaxy. Less than 1% of all galaxies are polar ring galaxies, making their data very limited. Manual classification of these galaxies can take months to years, and the human error rate is high. For this re ason, polar ring galaxies have not been the focus of astronomical research. We present a method to classify polar ring galaxies using a convolutional neural network (CNN). Multiclass classification was utilized, where the categories are smooth galaxies, disk galaxies, and polar ring galaxies. A CNN was trained on polar ring and non-polar ring data and then tested on a real unclassified dataset of 7,840 NGC (New General Catalog) galaxies. Four polar ring galaxies, 426 smooth galaxies, and 412 disk galaxies w ere extracted from this previously unclassified dataset at a rate of 130 galaxies per minute. As upcoming sky surveys continue to be released—such as the WALLABY sky survey, which will contain candidates for polar ring galaxies — machine learning models such as the one we propose could be tremendously useful to better understand the mechanisms by which a plethora of galaxies form. Wyoming and Colorado National Ground-Level NO2 Predictions via Satellite Imagery Driven Hybrid Neural Networks Elton L. Cao Fairview High School, Boulder, CO Outdoor air pollution, specifically nitrogen dioxide (NO 2), poses a global health risk. Land use regression (LUR) models are widely used to estimate ground-level NO2 concentrations by describing the satellite land use characteristics of a given location using buffer distance averages of variables. However, information may be leaked in this approach. Therefore, in this study, I leverage a convolutional neural network ( CNN) architecture to directly pass pixel plots of satellite imagery for the prediction of U.S. national ground -level NO2. I designed CNN architectures of various complexity which inputs both image and numerical based data, testing both high and low resolution pixel plots. My resulting model accurately predicted NO 2 concentrations at both daily (R 2 = 0.898) and annual (R 2 = 0.964) temporal scales, with coarse resolution imagery and simple CNN architectures displaying the best and most efficient performance. Furthermore, the CNN outperforms traditional buffer distance models, including random forest (RF) and neural network approaches. Additionally, with a novel graph neural network (GNN) based approach which leverages network interactions between monitoring sites, I developed a hybrid hierarchical GNN -CNN model which captures both short and long -distance associations between monitors. The resulting hybrid model significantly improved prediction against new and unseen monitors. With the success of hybrid neural networks in this approach, satellite land use variables continue to be useful for the prediction of NO2. Using this computationally inexpensive model, I encourage the globalization of advanced LUR models as a low - cost alternative to traditional NO2 monitoring.

Competition history

• JSHS 2024 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.