Detecting Stationary Atmospheric Waves in Venus with a Self-Supervised Adversarial Model Using Anomaly Detection
Published in Korea Software Congress (KSC), 2023
Recommended citation: Jun Seong Kim, Husnu Baris Baydargil, Jose Eduardo Silva, Yeon-Joo Lee, Meeyoung Cha. (2023). " Detecting Stationary Atmospheric Waves in Venus with a Self-Supervised Adversarial Model Using Anomaly Detection." Korea Software Congress. 1(1). https://jeon0001.github.io/junseongkim.com/files/VenusPaper2.pdf
Awarded Grand Prize for best student paper (최우수상)
Abstract
Atmospheric waves on Venus are formed due to temperature changes, pressure, and other atmospheric variables propagating through the atmosphere. These patterns and oscillations may provide valuable insights into the planet’s atmospheric dynamics and further characterization of the phenomenon. In this paper, we approach the detection of these atmospheric waves through an adversarially trained self-supervised anomaly detection model. We investigate two types of imaging: longwave infrared (LIR) and ultraviolet imaging (UVI). Distinct characteristics of LIR and UVI data present new opportunities to analyze the phenomenon and compare the machine learning performance. The results show that the model can differentiate between stationary waves and cloud formations, with an AUC score of 86.53% for LIR images and 90.81% for UVI images. Moreover, anomaly scores show UVI data is much less prone to mislabeling for normal cloud formations and stationary waves.