Automated lunar age detection through determining the day of the synodic month using convolutional neural networks

This paper introduces an image classification-based technique for determining the moon's age in a synodic month from a lunar picture. Before this study, no image classification-based approach has been applied for this purpose. Historically, mathematical methods have been employed for this determination, dating back to the Babylonian era. Our approach utilizes CNNs (Convolutional Neural Networks) to analyze lunar images. We utilized a pre-trained ResNet18 architecture for transfer learning, leveraging its pre-trained weights to adapt to the new lunar dataset. The dataset was collected from Nasa's moon phase and liberation website using the requests and selenium libraries. Finally, we trained and evaluated different models using accuracy metrics and confusion matrices. While experimenting in different experimental setups, the highest accuracy obtained in our study is 82.74%.