Spectral and Morphological Classification of Celestial Objects using Physics Informed Machine Learning

The classification of celestial objects has traditionally relied on spectral analysis, which involves identifying unique patterns in the light emitted or absorbed by the objects. In this study, we present a novel approach to categorizing celestial objects based on their spectral properties and morphological structures, leveraging physics-informed machine learning (PIML) techniques to enhance the classification accuracy and efficiency. In the spectral analysis, using Sloan Digital Sky Survey (SDSS) dataset, we classified stars, galaxies, and quasars, focusing on redshift and near-infrared values with i and z filters. Our analysis employed three machine learning models (K-Neighbors, SVM, Random Forest), and the performances were verified to be resilient applying hyperparameter tuning and cross-validation. The best-performing model, Random Forest Classifier, achieved 98% accuracy. In the morphological analysis, the YOLO v5 model was utilized on a customized dataset of astronomical images to further classify the Galaxy Class into five distinct categories and the YOLOv5 model detected those with an average precision of 92.6% and recall of 68.9%. In addition, the interpretability of the model is enhanced by the incorporation of physical knowledge, allowing astronomers to gain deeper insights into the underlying processes. Integrating physical knowledge into these ML models enhances interpretability, facilitating automated real-time analysis of big astronomical data and advancing our understanding of the cosmos.