Computer-Aided Mapping of Hyper- and Multi-Spectral Data

• Applicants: Professor Dr.-Ing. Gernot A. Fink; Professor Dr. Christian Wöhler
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2015 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 269661170

• Final Report Year: 2023

Final Report Abstract

The project aimed at the mapping of extraterrestrial surfaces using multi- and hyperspectral data. By classification and weakly supervised semantic segmentation of various surface spectra, regions could be grouped based on different properties. The developed methods were qualitatively and quantitatively evaluated and showed promising results. The main focus was on investigating extraterrestrial surfaces, particularly those of Mars and the Moon, as well as Earth. The automatic classification and mapping could provide important information about geological structures, land use, and land cover. This can be used for the geological mapping of entire planets and significantly expedite the search for interesting landing sites, as well as automatically finding comparable geological structures on planetary surfaces. On Earth, the methods can be used to map landuse classes on a large scale with minimal annotations. The developed self-supervised methods can be utilized to observe seasonal effects for various objects and areas. This can be helpful, for example, in monitoring long-term climate changes. The results of the project also led to further applications and insights in lunar remote sensing. Through automated crater detection combined with a diffusion model, it was demonstrated that specific volcanic structures on the lunar surface are likely much younger than previously assumed. Furthermore, several datasets were published to aid in understanding geological structures, landuse classes on Earth, and seasonal effects, promoting comparability of the developed methods. Overall, the results of this project contribute to improving the understanding and exploration of extraterrestrial and terrestrial surfaces, developing new machine vision methods for remote sensing, and supporting practical applications.

Publications

• Unsupervised Learning of Scene Categories on the Lunar Surface. Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, 614-621.
  Wilhelm, Thorsten; Grzeszick, Rene; Fink, Gernot & Wöhler, Christian
  
• DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars. Remote Sensing, 12(23), 3981.
  Wilhelm, Thorsten; Geis, Melina; Püttschneider, Jens; Sievernich, Timo; Weber, Tobias; Wohlfarth, Kay & Wöhler, Christian
  
• Generation of Attributes for Highly Imbalanced Land Cover Data. 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2616-2619.
  Kobmann, Dominik; Wilhelm, Thorsten & Fink, Gernot A.
  
• Land Cover Classification from a Mapping Perspective: Pixelwise Supervision in the Deep Learning Era. 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2496-2499.
  Wilhelm, Thorsten & Kossmann, Dominik
  
• Towards Tackling Multi-Label Imbalances in Remote Sensing Imagery. 2020 25th International Conference on Pattern Recognition (ICPR), 5782-5789.
  Kossmann, Dominik; Wilhelm, Thorsten & Fink, Gernot A.
  
• Uncertainty Guided Recognition of Tiny Craters on the Moon. 2020 25th International Conference on Pattern Recognition (ICPR), 5198-5205.
  Wilhelm, Thorsten & Wohler, Christian
  
• Uncertainty Introduced by Darkening Agents in the Lunar Regolith: An Unmixing Perspective. Remote Sensing, 13(22), 4702.
  Hess, Marcel; Wilhelm, Thorsten; Wöhler, Christian & Wohlfarth, Kay
  
• “DoMars24k: Expanding Automated Geomorphic Analysis on Mars by Wind and Ice Shaped Landforms”. In: 52nd Lunar and Planetary Science Conference. Lunar and Planetary Science Conference. 2021, 1901, S. 1901
  T. Wilhelm; R. Nocon; S. Stepcenkov & C. Wöhler
  
• Image Augmentations in Planetary Science: Implications in Self-Supervised Learning and Weakly-Supervised Segmentation on Mars. 2022 26th International Conference on Pattern Recognition (ICPR), 2800-2806.
  Kosmann, Dominik; Matei, Arthur; Wilhelm, Thorsten & Fink, Gernot A.
  
• Learning the Link between Albedo and Reflectance: Machine Learning-Based Prediction of Hyperspectral Bands from CTX Images. Remote Sensing, 14(14), 3457.
  Stepcenkov, Sergej; Wilhelm, Thorsten & Wöhler, Christian
  
• Machine Learning on Mars: Open Challenges, Similarities and Differences to Earth Remote Sensing. IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 5373-5376.
  Wilhelm, Thorsten; Kobmann, Dominik & Wohler, Christian
  
• “A Tomographic View into the Layering of the Lunar Regolith”. In: LPI Contributions 2678 (2022), S. 1497
  C Wöhler; T Wilhelm; R Bugiolacchi & S Althoff