Vous êtes ici : Accueil » Kiosque » Annonce

Identification

Identifiant: 
Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?

Annonce

24 mars 2017

Consideration of the underwater environment in the process of Automatic Target Recognition


Catégorie : Doctorant


Host laboratory: Lab-STICC (CNRS UMR 6285)

Advisors: I. Quidu, G. Le Chenadec and A. Baussard (Supervisor)

Dead Line : 28/04/2017

 

Motivations

For the purpose of underwater mine countermeasure, automatic (underwater) target recognition (ATR) is usually performed on high resolution sonar images. However, environmental effects are known to degrade performances of most of the existing ATR processes. Up to now, these effects are usually limited by filtering the image but it must be noticed that this strategy is not relevant for some environmental conditions and can also alter the available information [1, 2, 3].

In this thesis, on the contrary, we consider the environment as an important piece of information that can be used to design a robust ATR process.The challenge is to process sidescan sonar images to represent and to quantify information of the seafloor nearby a target in order to help the detection and the recognition stages. Previous works at ENSTA Bretagne consisted in describing the seafloor by the use of a new mathematical operator: the monogenic signal [4]. Combined with the intrinsic dimensions in a multi-scale framework, this operator allows to characterize the seafloor in terms of energetic and geometrical properties at several levels of details [5].

Expected Work

Following the above mentioned works on seafloor characterization, the candidate will have to analyze and then to include this information in the ATR process so as to improve ATR performances.

Analysis means understanding, processing and evaluating the available mathematical tools developed in an ongoing PhD thesis [6, 7, 8]. The recently proposed classification of seafloor as homogeneous, anisotropic and complex will be assessed and discussed. It would help the candidate to build a robust ATR process. Whereas the standard ATR process is sequential and likely to propagate errors, the candidate will have to propose a new adjustable framework in which environment properties can be set as inputs for adjusting the target recognition algorithm.

Contact

alexandre.baussard@ensta-bretagne.fr

 

Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2015.