IMT ATLANTIQUE - Deep learning for a realistic simulation of degradation in underwater imagery [postdoc]

Posdoctoral position (18 months) in deep learning for a realistic simulation of degradation in underwater imagery

 

Context

IMT-Atlantique, an engineering school under the supervision of the Ministry of Industry, is looking for a postdoctoral fellow for 18 months, to start as soon as possible as of September 2020. The position is based on Brest campus of the school. The candidate will join the Signal and Communications department, within the TOMS team of the Lab-STICC (Traitement, Observations et Méthodes Statistiques), whose research activities include signal and image processing for remote sensing, and learning dynamical systems using arti_cial intelligence.

The open position is part of the SUBSEE4D project co-sponsored by Cervval, a company specializing in digital simulation and decision support in complex systems, and France Energies Marines, a research institute for marine renewable energy (MRE) and IMT Atlantique. Installed or floating wind turbines and tidal turbines are major infrastructure projects that are costly in terms of investment and also, due to the restrictive marine environment, in terms of operation and maintenance, particularly for the submerged part: inspection of anodes, corrosion detection, monitoring of organic aggregates (bio-colonisation), monitoring of the structural condition, particularly of underwater cables and anchor lines. MRE eet managers will face unprecedented operating and maintenance challenges, even greater than those faced by offshore oil and gas companies, due in particular to the shallow waters that characterize MRE facilities: currents, waves, water turbidity and the proliferation of living organisms on the infrastructure make it difficult for divers and ROVs (Remote Operated Vehicles) to operate and maintain the facilities, and degrade and damage the infrastructure.

Within this framework, the objective of this position is to develop an image degradation generator via AI approaches, allowing the implementation of a real-time simulation of underwater vision conditions with different parameters controlling the effect of different turbidities. Indeed, a conventional simulator requires powerful processor resources to provide the ability to simulate a degraded environment and for the creation of image databases. To circumvent the difficulty and cost of acquiring underwater images, we will use synthetic images whose degradation is carried out by a specialized IFREMER simulator. State-of-the-art generative model learning techniques such as Generative Adversarial Networks (GANs) will be used to generate realistic data from these databases.

The candidate will have to define and develop and validate generative model architectures adapted to the problem of shallow water degradation simulations. The work will be valorized through methodological and applied publications in international conferences and scientific journals in the field.

 

Candidate

The candidate must have a PhD in image or signal processing, machine learning or related _elds, or equivalent experience. Qualifications required:

  • Machine Learning, Signal and Image Processing, Applied Mathematics
  • Programming in Python
  • Experience in deep learning methods, including generative models, and associated libraries in Python (Keras, Tensorow, Pytorch)
  • Interest in Underwater Imagery and computer vision

 

Contact

Lucas Drumetz, Associate Professor, IMT-Atlantique, Signal and Communications Department, and UMR LabSTICC, TOMS team. Please send a resume and cover letter to Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser..

© 2020 BreTel