Saliency Attention and SIFT Keypoints Combination for Automatic Target Recognition on MSTAR dataset

Ayoub Karine (1) , Abdelmalek Toumi (2, 3) , Ali Khenchaf (4, 3) , Mohammed El Hassouni (1)

1 LRIT - Laboratoire de Recherche Informatique et Télécommunications
2 Lab-STICC_ENSTAB_CID_TOMS
3 Pôle STIC_REMS
4 Lab-STICC_ENSTAB_MOM_PIM

This paper aims to present a novel method for automatic target recognition based on synthetic aperture radar (SAR) images. In order to describe a region of interest (target area), we use a saliency attention model. Then, the produced saliency map is used as a mask on SAR image in order to separate the ground target from the background. After that, we calculate the scale invariant feature transform (SIFT) descriptors of the transformed SAR image. In this way, we maintain only the SIFT keypoints located in the salient region. This strategy leads not only to reduce the dimensionality but also enhances its discriminative power. For recognition step, a matching approach between vector descriptors of unknown image target and all known images stored in training data set is adopted. To validate the proposed approach, MSTAR data set is used. The obtained experimental results show that our approach can effectively describe a SAR image, and obviously improve the recognition rate.

Mots clés: Automatic target recognition; synthetic aperture radar; SIFT; saliency attention model; matching

Domaines: Intelligence artificielle [cs.AI]; Traitement du signal et de l'image [eess.SP]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]

Ayoub Karine, Abdelmalek Toumi, Ali Khenchaf, Mohammed El Hassouni. Saliency Attention and SIFT Keypoints Combination for Automatic Target Recognition on MSTAR dataset . 3rd International Conference on Advanced Technologies for Signal and Image Processing - ATSIP’2017, Mar 2017, fez Morocco. ⟨10.1109/ATSIP.2017.8075558⟩. ⟨hal-01657530⟩