Galaxy Morphology Classification Based on VGG19 Deep Convolutional Neural Network

Galaxy morphology classification is crucial for understanding galaxy formation and evolution. Traditional methods for classifying galaxy morphologies are labour-intensive and require expert knowledge. Recent deep learning approaches simplify this process but often focus only on basic galaxy types, such as spiral, elliptical, and irregular, overlooking more complex or mixed morphologies. This study applied a transfer learning-based model for 10-class galaxy morphology classification using the Galaxy10 DECaLS dataset, which contains 17,736 images across 10 distinct galaxy types that diverge from the primary galaxy morphology types. The proposed classification model employed VGG19 as the base for feature extraction, supplemented by custom fully connected layers for classification. The approach included comprehensive data preprocessing, extensive data augmentation, the use of learning rate reduction and model checkpointing during training. A thorough evaluation using multiple metrics demonstrated significant improvements in classification performance, achieving 97.97% accuracy, 89.84% precision, 89.69% recall, an F1-score of 89.68%, and an area under the curve (AUC) of 0.98 on test data using macro-averaging. These findings confirmed the efficacy of deep learning models for automating galaxy morphology classification and underscored the benefits of transfer learning in achieving high accuracy.