Detection and Feature Analysis of Martian Dust Devil Based on Multimodel Deep Learning

Dust devils, common atmospheric phenomena on Mars, play a crucial role in Martian atmospheric dynamics. Previous studies reveal their prevalence across seasons and latitudes, particularly during summer and spring due to solar radiation influence. However, existing investigations have primarily focused on regional or short-term studies, leaving a gap in our understanding of dust devils distribution on a global scale and their behavior over long-term cycles. This work introduces the Martian Dust Devils Ultra (MDD-Ultra) approach for automated identification and three-dimensional scale estimation of Martian dust devils. The method employs the Martian Dust Devil Detection-Cascade algorithm (MDDD-Cascade) for enhanced dust devil recognition, Martian Dust Devil Semantic Segmentation algorithm (MDDSS) for precise segmentation, and Martian Dust Devil Three-Dimensional (MDD-3D) algorithm for morphological parameter calculation. Analysis of 583 dust devils using the Context Camera (CTX) images dataset from the Mars Year 36 demonstrated improved detection accuracy and efficiency compared to existing algorithms. Spatially, dust devils are concentrated in the northern Amazonis Planitia and southern Hellas Planitia regions, showing an east-west distribution pattern. The study establishes a log-linear relationship between dust devil radius and height, shedding light on their distribution and morphology. This research not only advanced automated dust devil analysis but also provided valuable insights for future Mars exploration, aiding in rover landing site selection and safety assessment.