The detection of astronomical transient events—such as supernovae, gamma-ray bursts, and stellar flares—has become increasingly vital in astrophysics due to their association with extreme cosmic processes. However, identifying these short-lived phenomena within massive sky survey datasets, like those from the GOTO project, poses major challenges for traditional analysis methods. This study proposes a Deep Learning approach using Convolutional Neural Networks (CNNs) to improve transient classification. By leveraging Transfer Learning and Fine-Tuning on pre-trained ImageNet models, the system adapts to the specific features of astronomical imagery. Data Augmentation techniques—including rotation, flipping, and noise injection—are employed to enhance dataset diversity, while Dropout and varying Batch Sizes are explored to reduce overfitting and improve generalization. To further boost performance, Ensemble Learning strategies such as Soft Voting and Weighted Voting are implemented, combining multiple CNN models for more robust predictions. The results demonstrate that this integrated approach significantly enhances the accuracy and reliability of transient detection, offering a scalable solution for real-time applications in large-scale surveys like GOTO.
Prommool et al. (Tue,) studied this question.