Leveraging Movement Representation from Contrastive Learning for Asteroid Detection

Abstract

Asteroid detection is a critical task in astronomy for planetary defense and understanding the solar system’s evolution. Traditional methods often struggle with low signal-to-noise ratios and the vast amount of data generated by modern sky surveys. This study introduces a novel deep learning approach that leverages movement representation learned through contrastive learning to improve the accuracy and efficiency of asteroid detection. By training the model to recognize the specific spatiotemporal patterns of moving objects against a static stellar background, the proposed framework enhances the ability to distinguish true asteroids from artifacts and noise. The methodology was evaluated using data from the Gravitational-wave Optical Transient Observer (GOTO) telescope, demonstrating significant improvements in detection performance. This research highlights the potential of self-supervised representation learning in automating astronomical discoveries and handling the data challenges of next-generation surveys.