Plug-and-Play for synthetic aperture radar

19 Novembre 2024

---

Catégorie : Stagiaire

---

The full subject, including references and figures, can be found at https://partage.imt.fr/index.php/s/MCwCbpRewPQCwsk

Context

The remote sensing field aims at exploiting satellite or aerial images for earth observation. This, in turns, has an increasingly important social impact, as the applications fo these methods include tackling key challenges such as environment monitoring (e.g. urban monitoring, deforestation, crop and agriculture monitoring...). This internship specifically focuses on remote sensing Synthetic Aperture Radar (SAR) images. Such images are highly useful since, due to the active nature of the sensor, it is possible to acquire them even during night or in the presence of clouds. They are generated by recording the backpropagation of an emitted pulse signal; in the different image pixels, this signal can either interact with diffuse scatterers (e.g. vegetation), generating the image background, or with strong backscatterers (e.g. building corner) generating cardinal sines -- called targets -- in the image, due to the sensor Point Spread Function (PSF).

Background / target separation

In some applications, it is useful to localize and separate the targets from the SAR image. From a mathematical point of view, this amounts to decompose the considered SAR image x into: x = n + t, with n the image background and t the target map. Nevertheless, performing such a task is complicated due to the fact that strong backscatterers have a continuous position in the scene, while the SAR image has a discrete nature due to the sampling process. Consequently, finding the targets in the image requires to localize them at a sub-pixel precision. This calls for additional priors to solve the problem.

We already achieved good results on simulated 1-dimensional dataset by leveraging a handcrafted sparsity prior. Nevertheless, such an approach tends to give disappointing results on 2-dimensional real images, and we would like to go a step further by incorporating data-driven regularizations within our algorithm. To do so, we propose to adopt the Plug-and-Play paradigm, in which a deep denoiser network is plugged within an iterative algorithm. Specifically, we intend to use the state-of-the art self-supervised SAR denoising method named MERLIN.

Candidates

The candidate must be a Master 2 student (or equivalent) with a good knowledge of signal/image processing and machine learning. Ideally, the candidate will be familiar with the Python language (and in particular with pytorch).

The candidate will acquire an expertise in remote sensing, optimization, deep learning and self-supervised training strategies. Such competences are largely sought-after, both in academia and in private companies, and are transferable to many other applications such as biomedical imaging, astrophysics... In addition, if relevant, the work conducted during the internship might lead to a scientific publication.

An extension of this subject to a PhD might be considered.

Contact

The internship (6 months) will take place in the IMAGES team (Télécom Paris) under the supervision of Christophe Kervazo and Saïd Ladjal.

- Contact: christophe.kervazo@telecom-paris.fr

- More information on: https://sites.google.com/view/christophekervazo/