Context
While coastal cliffs represent about 52% of the global coastline (Young and Carilli, 2019), cliff erosion is likely to increase as sea level rises, endangering surrounding populations and infrastructures (R. Cooley et al., 2022). Bringing new challenges to research, climate change calls for the development of innovative methods for Earth observation and monitoring. This research topic is part of the HIRACLES (CNES) project (https://www-iuem.univ-brest.fr/pops/projects/hiracles?jump=welcome) which aims to develop a new optimized approach to detect and quantify cliff front erosion using Pl ́eiades imagery. The master thesis will be carried out in connection with the CICERO doctoral thesis (Contribution of multiangular spatial imagery to the monitoring and understanding of cliff erosion), started in 2021.
The CICERO doctoral thesis intends to:
- Test and adapt existing deep learning (DL) change detection methods (Siamese change detection networks), taking as input pairs of oblique satellite images (Pléiades), to detect cliff erosion through time series;
- Apply a 3D reconstruction of the cliff face over previous erosion areas detected by artificial intelligence (AI).
- Analyze relationships between retreat rates, volumes of rockfalls and environmental factors on the various studied sites.
As part of this doctoral thesis, the proposed master thesis will focus on developing a new AI method, taking as input a single image acquired from oblique satellite imagery (Pléiades, Pléiades Néo) or drone imagery. Thus, a comparison between two deep learning approaches can be conducted:
- multi-temporal detection through change detection methods on image pairs (CICERO doctoral thesis);
- mono-temporal detection through change detection methods on single images (master thesis).
Objectives
The main objective of this master thesis is to develop a methodology to provide a coarse delineation of rockfalls in a dataset made of oblique images. The originality comes from the fact it will be achieved without using multi-date image pairs comparison (i.e. no Siamese neural networks). Several detection scales can be tested (leading to patch-based, pixel-based or object-based approaches).
Tasks
The work to be conducted within this master thesis is organized in 4 steps:
- Detection of erosion areas using DL methods on mono-date oblique images (no need of stereoscopic images) with supervision (Li and Hsu, 2020);
- Detection of erosion areas using DL methods on stereoscopic images with supervision: how multi-view scenes can help to detect erosion? (Chen et al., 2020; Sun et al., 2020; Reading et al., 2021)
- Detection of erosion areas using DL methods with low supervision: from an anomaly detection and localization perspective, the algorithm learns on samples without rockfalls and is able to detect when the input does not match (when the sample has rockfalls) (Shao et al., 2022). These anomalies are then considered as a rockfall;
- (Bonus) Retrieve the digital surface model (DSM) of the scene from a mono-date oblique image (Panagiotou et al., 2020).
The objectives may be adjusted according to the results at each step and the student’s abilities.
Organization
The daily supervision will be provided mainly by Zoé Bessin (Geo-Ocean and LETG Brest, CICERO doctoral student, zoe.bessin@univ-brest.fr), completed with the support of senior scientists: Sébastien Lefèvre (IRISA, Vannes, sebastien.lefevre@irisa.fr). The work will be done in Brest at LETG laboratory, Technopole.
Profile
- Knowledge in deep learning algorithms
- Object detection
- Python, PyTorch
- Remote sensing
- Interest in coastal issues and geomorphology
References
Y. Chen, S. Liu, X. Shen, and J. Jia. DSGN: Deep Stereo Geometry Network for 3D Object Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12533–12542, Seattle, WA, USA, 2020. IEEE. ISBN 978-1-72817-168-5. URL https://doi.org/10.1109/CVPR42600.2020.01255.
W. Li and C.-Y. Hsu. Automated terrain feature identification from remote sensing imagery: a deep learning approach. International Journal of Geographical Information Science, 34(4):637–660, 2020. ISSN 1365-8816. URL https://doi.org/10.1080/13658816.2018.1542697.
E. Panagiotou, G. Chochlakis, L. Grammatikopoulos, and E. Charou. Generating Elevation Surface from a Single RGB Remotely Sensed Image Using Deep Learning. Remote Sensing, 12(12):2002, 2020. ISSN 2072-4292. URL https://doi.org/10.3390/rs12122002.
S. R. Cooley, D. S. Schoeman, L. Bopp, P. Boyd, S. Donner, S.-I. Ito, W. Kiessling, P. Martinetto, E. Ojea, M.-F. Racault, B. Rost, and M. Skern-Mauritzen. 2022: Oceans and Coastal Ecosystems and Their Services. In Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, chapter 3, pages 379–550. Cambridge University Press, Cambridge, UK and New York, NY, USA, 2022. doi: 10.1017/9781009325844.005. URL https://epic.awi.de/id/eprint/56137/1/IPCC_AR6_WGII_Chapter03.pdf.
C. Reading, A. Harakeh, J. Chae, and S. L. Waslander. Categorical Depth Distribution Network for Monocular 3D Object Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8551–8560, Nashville, TN, USA, 2021. IEEE. ISBN 978-1-66544-509-2. URL https://doi.org/10.1109/CVPR46437.2021.00845.
F. Shao, L. Chen, J. Shao, W. Ji, S. Xiao, L. Ye, Y. Zhuang, and J. Xiao. Deep Learning for Weakly-Supervised Object Detection and Localization: A Survey. Neurocomputing, 496:192–207, 2022. ISSN 09252312. URL https://doi.org/10.1016/j.neucom.2022.01.095.
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou, and H. Bao. Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10545–10554, Seattle, WA, USA, 2020. IEEE. ISBN 978-1-72817-168-5. URL https://doi.org/10.1109/CVPR42600.2020.01056.
A. P. Young and J. E. Carilli. Global distribution of coastal cliffs. Earth Surface Processes and Landforms, 44(6):1309–1316, 2019. URL https://doi.org/10.1002/esp.4574.