Semantics-Assisted Multiview Fusion for SAR Automatic Target Recognition

Multiview fusion algorithms have been widely applied in synthetic aperture radar automatic target recognition (SAR ATR). However, they usually ignore semantic information, which results in limited recognition performance. Therefore, how to effectively integrate semantic information into multiview fusion algorithms and explore the correlation between multiple views and semantic information is a crucial and urgent problem. To address this problem, we develop a semantics-assisted multiview fusion (SMVF) algorithm, which includes three loss terms, i.e., view-specific loss term, semantics-regularized loss term, and view-semantics-coupled loss term. To be specific, the view-specific and semantics-regularized loss terms convert multiple views and semantic information into their corresponding sparse codes, respectively. The view-semantics-coupled loss term constrains the sparse codes of multiple views and semantic information to explore their correlation. Finally, these three loss terms are jointly optimized to acquire the optimal sparse codes to calculate the reconstruction errors for recognition, by which SMVF not only effectively exploits semantic information, but also explores the correlation between multiple views and semantic information. Extensive experiments demonstrate that SMVF achieves high recognition accuracies (i.e., 99.7%, 91.6%, and 97.2%) under three scenarios (i.e., EOC-1, EOC-2, and SAR-ACD), which are better than other advanced algorithms.