Skip to Main content Skip to Navigation

Traitement et classification parcimonieuse des images radar pour l’aide à la reconnaissance de cibles

Abstract : Automatic target recognition has become a flourishing research topic in remote sensing. This problematic is of paramount importance in several military and civilian applications (security, surveillance, automobile, environment, medicine, ...). In this work, we focus on the development of new methodology dedicated to the target recognition from synthetic aperture radar images (inverse or direct). In this context, the different methods proposed in the literature have several drawbacks according to the type of used data (heterogeneous, multimodal, …), the accuracy, the robustness to noise and the computation time. In this work, we aim to propose new methods for targets recognition from radar images. Thus, two different databases are considered. On the one hand, we use the ISAR (Inverse Synthetic Aperture Radar) images acquired from the anechoic chamber of ENSTA Bretagne. On the other hand, we exploit the SAR (Synthetic Aperture Radar) images of MSTAR database. To achieve the goal of the amelioration of the recognition process studied and developed, we give a special interest to the SRC (Sparse Representation-based Classification) method to recognize the radar images. This method includes also the processing and feature extraction steps to build the dictionary. In this optic, the first contribution consists on constructing a new dictionary composed by the SIFT (Scale Invariant Feature Transform) descriptors filtered by the saliency attention method. After that, this dictionary is exploited by a multitask sparse classifier to achieve the recognition step. In the second contribution, we statistically model the radar images in the complex wavelet domain. The resulting statistical parameters (univariate or multivariate) are stacked together to construct the statistical dictionary. Afterwards, this dictionary is weighted by using a similarity measure that includes the KLD (Kullbak-Leibler Divergence) between the statistical parameters. The performances of the two proposed methods have been evaluated empirically on two different databases of radar images (ISAR and SAR).
Complete list of metadatas

Cited literature [215 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, April 1, 2020 - 3:49:14 PM
Last modification on : Tuesday, April 7, 2020 - 3:18:27 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02096884, version 2


Ayoub Karine. Traitement et classification parcimonieuse des images radar pour l’aide à la reconnaissance de cibles. Traitement du signal et de l'image [eess.SP]. ENSTA Bretagne - École nationale supérieure de techniques avancées Bretagne; Université Mohammed V (Rabat). Faculté des sciences, 2018. Français. ⟨NNT : 2018ENTA0013⟩. ⟨tel-02096884v2⟩



Record views


Files downloads