Exoplanet Classification through Convolutional Vision Transformers with Image-Based Transit Analysis

Detecting exoplanet candidates from photometric time series is a critical task in modern astrophysics, often hindered by noisy signals, class imbalance, and subtle transit features. In this study, we propose a novel approach based on the Convolutional Vision Transformer (CvT) to classify exoplanet transit events using two-dimensional representations derived from light curve data. The CvT architecture combines convolutional token embeddings with hierarchical self-attention, enabling the model to capture both local transit patterns and global contextual dependencies. Our model demonstrates strong classification performance, achieving an accuracy of ۸۷% and an Area Under the ROC Curve (AUC) of ۰.۹۴۳ on a test set constructed from real transit data. The confusion matrix analysis highlights a high true positive rate (۹۳.۹%), indicating that the model is highly sensitive to planetary transits while maintaining reasonable control over false positives. Compared to prior work based on pure Vision Transformers, our CvT-based model provides superior spatial feature extraction with fewer parameters.

کلیدواژه ها:

Convolutional Vision Transformers ، Exoplanet Classification ، Recurrence Plot