Tuesday 11 March 2025
A team of researchers has developed a novel approach for classifying galaxy profiles using machine learning techniques and convolutional neural networks (CNNs). The methodology, which combines unsupervised learning methods with CNNs, has been successfully applied to analyze a large dataset of neutral atomic hydrogen (HI) spectral lines.
The study focused on the HI line, which is a crucial tool for understanding galaxy evolution and dynamics. By analyzing the shape and properties of this spectral line, astronomers can gain insights into the gas kinematics within galaxies, star formation processes, and interactions between galaxies. However, with the increasing volume and complexity of radio astronomy data, traditional methods for analyzing these profiles have become time-consuming and limited.
The researchers used a dataset of 318 HI spectra from the Catalogue of Isolated Galaxies (CIG) and 30,780 spectra from the Arecibo Legacy Fast ALFA Survey (ALFALFA). They first applied unsupervised clustering methods to identify patterns in the data, followed by iterative fitting with polynomial, Gaussian, and Lorentzian models. The resulting profiles were then classified using a range of machine learning algorithms, including K-Nearest Neighbors, Support Vector Machines, and Random Forest.
To further improve classification accuracy, the researchers introduced an additional dimension to the HI profiles by creating 2D images from the original data. These images were generated by rotating, subtracting, or normalizing the spectra to highlight specific features. The resulting classifications showed a significant improvement in accuracy compared to traditional 1D methods.
The study’s results have important implications for future radio astronomy surveys, such as the Square Kilometre Array (SKA), which is expected to produce millions of HI profiles. By developing efficient and accurate classification methods, researchers can quickly identify patterns and anomalies in these large datasets, allowing them to focus on specific galaxies or phenomena.
The methodology developed by this team has been made publicly available through a GitHub repository, providing an open-access platform for other researchers to reproduce the results and adapt the approach to their own research. The study’s findings demonstrate the potential of machine learning techniques to enhance our understanding of galaxy evolution and dynamics, and highlights the importance of collaboration between astronomers and computer scientists.
The researchers’ approach has also shed light on the challenges of working with large-scale astronomical datasets. The study found that computational resources are a critical limitation for these types of analyses, emphasizing the need for high-performance computing facilities and optimized algorithms to process these data efficiently.
Cite this article: “Classifying Galaxy Profiles with Machine Learning Techniques”, The Science Archive, 2025.
Galaxy Profiles, Machine Learning, Convolutional Neural Networks, Unsupervised Learning, Spectral Lines, Radio Astronomy, Galaxy Evolution, Dynamics, High-Performance Computing, Large-Scale Datasets
