Speaker
Description
Airglow phenomena serve as essential tracers for understanding the dynamics and interactions within the Earth's upper atmosphere. In this study, we introduce a machine learning-based approach for classifying airglow images captured by the all-sky camera at the Shigaraki (SGK) observatory. Our objective is to automate the identification of gravity wave signatures within these images, distinguishing them from those without such features. To achieve this, we developed a supervised Convolutional Neural Network (CNN) model trained on an extensive dataset spanning 11 years (2009–2019) (Solar cycle 24),with images annotated by domain experts. The methodology includes rigorous preprocessing, data augmentation (e.g., rotation and noise injection), and the application of advanced deep learning algorithms to improve classification performance and reduce overfitting. The trained model achieved 98.7% accuracy in distinguishing gravity wave features from non-wave images,validated through 5-fold cross-validation and benchmarked against a Support Vector Machine (SVM) baseline. Our results demonstrate the effectiveness of deep learning in enhancing gravity wave detection, offering a scalable and efficient solution for analyzing large volumes of airglow imagery. This study highlights the potential of artificial intelligence in atmospheric research,paving the way for improved monitoring and understanding of upper atmospheric dynamics.
