Unconventional applications for geo-spatial deep learning

https://www.replicaswatches.org/ exact replica watches www.shoponlinewatches.com tag heuer replica fake watches rolex replica watches kochamzegarki.pl fake rolex replique montre replica relojes replica watches rolex replika imitasyon saat

Unconventional applications for geo-spatial deep learning

Keywords: deep learning, explainability, interpretability, on-board processing, heritage mapping, biodiversity, SAR, optical, hyperspectral, LiDAR

Session abstract
Since the XXIII ISPRS Congress in Prague (2016), deep learning methods have emerged and proved very successful in providing meaningful insights from big archives of Earth Observation data. They are now regarded as the new state of the art in remote sensing data analysis and are leading to breakthroughs e.g. in land cover / use classification, change detection or data fusion. A great number of deep learning publications in photogrammetry and remote sensing journals and conferences focus on the most classical applications such as supervised classification. Most “low-hanging fruits” have been harvested already.

Deep learning is explicitly addressed in two ISPRS working groups, WG II/6: Large-scale Machine Learning for Geospatial Data Analysis and WG III/4: Hyperspectral Image Processing. It is
however such a transversal topic that novel and relevant applications can be found within many working groups across ISPRS Technical Commissions I to IV. For this reason, we propose a
session about brave talks in new, inspiring applications of deep learning. In this thematic session, the selected presentations focus on novel, unconventional and ambitious
applications of deep learning, in various application domains of photogrammetry and remote sensing. Highlights of the session include three presentations on the interpretability / explainability
of deep learning models (a common pitfall or criticism of such models), and two presentations on adapting deep learning models to on-board processing on orbital platforms. A variety of themes is
covered such as biodiversity or heritage mapping, and the main geo-spatial image sources are represented (space-borne optical and SAR, aerial images, hyperspectral, LiDAR).

Tentative list of full papers/authors

1. Jan Dirk Wegner (ETH Zürich) – Deep learning for biodiversity estimation
2. Clément Mallet (IGN) – Transferability of deep-based models for aerial historical images
3. Fabio Remondino (Bruno Kessler Foundation FBK) – 3D heritage classification with artificial intelligence methods
4. Lewis Smith and Yarin Gal (University of Oxford) – Flood detection on low cost orbital hardware
5. François de Vieilleville (Agenium Space) and Stéphane May (CNES) – Learning from models: insights for simplification and on-board processing
6. Ronan Fablet, Lucas Drumetz and François Rousseau (IMT-Atlantique) – Learning energy-based representations with missing data
7. Ribana Roscher (University of Bonn) – Transparency to explainability: machine learning for the natural sciences
8. Mihai Datcu (DLR) – Explainable deep learning for SAR images
9. Devis Tuia (Wagenigen University) – Learning with deep models by asking the right questions
10. Matthieu Molinier (VTT Technical Research Centre of Finland) – The inadequacy of popular small scale HSI “benchmark” datasets for meaningful evaluation of deep learning models : theoretical and experimental proofs
Reserve list – in case of cancellations
11. Cheng Wang (Xiamen University / Fujian Key Lab), Jonathan Li (University of Waterloo) – LO- Net: Deep Real-time Lidar Odometry
12. Xiaoxiang Zhu (DLR / Technical University of Munich) – AI4EO: Reasoning, Uncertainty, and Ethics