Optical-radar fusion for rip-channel detection from Sentinel data

Context
The internship position is opened in the context of the EVERGREEN team project - https://team.inria.fr/evergreen/ .
Our team is actively working on the design and implementation of cutting-edge machine learning techniques to effectively exploit heterogeneous and multi-temporal Earth observation data for agricultrual and enviromental applications.
The team, located in a multidisciplinary laboratory, has an active and stimulating environment with master, PhD and Post-doc students coming from different countries.
The internship will be funded by a project in collaboration with INRAE.
Assignment
Assignments :
With the help of members of the EVERGREEN team (Cassio F. Dantas and Dino Ienco), the intern will add Sentinel-1 SAR data to an existing Sentinel-2 dataset for rip-channel detection (BAISen2), which was developed within the context of an ongoing PhD thesis in the team (Christopher Jabea). The intern will then assess what SAR imagery brings to the detection of rip channels along the French Atlantic coast, using deep learning detectors based on the DETR architecture and optical-radar fusion.
For a better knowledge of the proposed research subject :
Rip channels are deep incisions in surf-zone sandbars associated with rip currents, the leading cause of drowning on beaches worldwide [1]. Our team built BAISen2 [2], an open dataset of 2,058 Sentinel-2 patches with 12,504 bounding boxes. It covers 42 sites along 350 km of the French Atlantic coast from 2015 to 2024.
Optical imagery depends on daylight and cloud cover, but SAR does not, and rip-current signatures have already been seen in spaceborne SAR imagery [3]. The internship will address three questions:
Are rip channels visible in Sentinel-1 imagery?
How does a detector trained on Sentinel-1 alone compare with one trained on Sentinel-2?
Does combining Sentinel-1 and Sentinel-2 improve detection [4]?
The team’s existing codebase for training and evaluating DETR-like detectors will be extended to optical–radar fusion.
[1] B. Castelle, T. Scott, R. W. Brander, R. J. McCarroll. “Rip current types, circulation and hazard.” Earth-Science Reviews, 163:1–21, 2016.
[2] C. Jabea, J. Turpin, C. F. Dantas, B. Castelle, I. Manighetti, D. Ienco. “BAISen2: A Sentinel-2 image dataset for rip channels detection along the French Atlantic coast.” Zenodo, 2026.
https://doi.org/10.5281/zenodo.21980065
[3] J. C. B. da Silva. “SAR observation of rip currents off the Portuguese coast.” In V. Barale et al. (eds.), Remote Sensing of the European Seas, pp. 399–410. Springer, 2008.
[4] A. Fuller, K. Millard, J. R. Green. “CROMA: Remote sensing representations with contrastive radar-optical masked autoencoders.” NeurIPS, 2023.
Collaboration :
The intern will work closely with Christopher Jabea, a PhD student in the EVERGREEN team who works on rip-channel detection from satellite imagery and is the lead author of the BAISen2 dataset
Main activities
Main activities :
Review the scientific literature on rip-channel and rip-current detection, SAR imagery and optical–radar fusion.
Collect and preprocess Sentinel-1 SAR data that matches the BAISen2 dataset (same sites and acquisition dates).
Train and evaluate DETR-like detectors on Sentinel-1 alone and compare them with detectors trained on Sentinel-2.
Design and implement optical–radar fusion strategies in PyTorch, building on the team’s existing codebase.
Report and summarise the results
Additional activities:
Contribute an extension of the BAISen2 dataset with Sentinel-1 data, to be released openly.
If the results allow, prepare a scientific paper for submission to an international journal or conference.
Skills
Technical skills and level required:
Advanced Python programming skills
Proficiency in deep learning with PyTorch
Background in machine learning, computer vision or remote sensing
Experience with object detection is appreciated
Experience with SAR or satellite data is a plus
Languages:
English: good proficiency
French: a plus but not necessary
Relational skills:
Ability to work both on your own and in a team
Communication skills, including presenting research results and working in an interdisciplinary team
Analytical thinking and problem solving
Curiosity and a continuous-learning mindset
Other valued skills:
Interest in Earth observation and coastal applications