Plug-and-Play Synthetic Aperture Radar Image Formation Using Deep Priors

Basit Öğe Kaydı
dc.authoridSALEEM, AMMAR/0000-0002-3885-9596en_US
dc.authoridAlver, Muhammed Burak/0000-0001-6089-3793en_US
dc.authoridÇetin, Mujdat/0000-0002-9824-1229en_US
dc.contributor.authorAlver, Muhammed Burak
dc.contributor.authorSaleem, Ammar
dc.contributor.authorÇetin, Mujdat
dc.date.accessioned2024-07-12T21:37:47Z
dc.date.available2024-07-12T21:37:47Z
dc.date.issued2021en_US
dc.department[Belirlenecek]en_US
dc.description.abstractThe reconstruction of synthetic aperture radar (SAR) images from phase history data is an ill-posed inverse problem which, in several lines of recent work, is solved by minimizing a cost function. Existing reconstruction methods use regularization to tackle the ill-posed nature of the imaging task. However, in general, these regularizers are either too simple to capture complex spatial patterns and can only promote fixed, predefined features, or lead to non-quadratic cost functions that are nontrivial to minimize. Recently emerging plug-and-play (PnP) priors technique is a flexible framework that allows forward models of imaging systems to be integrated with state-of-the-art regularizers. Inspired by this, we propose a novel PnP SAR image reconstruction framework for spotlight-mode SAR. SAR involves complex-valued reflectivities with spatial structure on the reflectivity magnitudes that can be learned and imposed as priors. This distinguishing aspect is formulated into our proposed framework. We demonstrate the use and effectiveness of a convolutional neural network (CNN) based prior model for the reconstruction of synthetic and real SAR scenes and compare the results with FFT-based, non-quadratic regularization-based, and dictionary learning-based reconstruction methods as well as a PnP framework with BM3D regularizer. Our results suggest that these deep priors enable the learning and incorporation of complicated spatial patterns more effectively than existing methods, and produce significantly improved images especially from limited observations.en_US
dc.description.sponsorshipScientific and Technological Research Council of Turkey (TUBITAK); Higher Education Comission (HEC), Pakistanen_US
dc.description.sponsorshipThis work was supported in part by The Scientific and Technological Research Council of Turkey (TUBITAK) through a Graduate Fellowship, and in part by Higher Education Comission (HEC), Pakistan. The associate editor coordinating the reviewof this manuscript and approving it for publication was Dr. Francesco Soldovieri.en_US
dc.identifier.doi10.1109/TCI.2020.3047473
dc.identifier.endpage57en_US
dc.identifier.issn2573-0436
dc.identifier.issn2333-9403
dc.identifier.startpage43en_US
dc.identifier.urihttps://doi.org/10.1109/TCI.2020.3047473
dc.identifier.urihttps://hdl.handle.net/20.500.12415/6941
dc.identifier.volume7en_US
dc.identifier.wosWOS:000611075800001en_US
dc.identifier.wosqualityQ2en_US
dc.indekslendigikaynakWeb of Science
dc.language.isoenen_US
dc.publisherIeee-Inst Electrical Electronics Engineers Incen_US
dc.relation.ispartofIeee Transactions on Computational Imagingen_US
dc.relation.publicationcategoryMakale - Uluslararası Hakemli Dergi - Kurum Öğretim Elemanıen_US
dc.rightsinfo:eu-repo/semantics/closedAccessen_US
dc.snmzKY04283
dc.subjectImagingen_US
dc.subjectSynthetic Aperture Radaren_US
dc.subjectRadar Polarimetryen_US
dc.subjectImage Reconstructionen_US
dc.subjectConvex Functionsen_US
dc.subjectRadar Imagingen_US
dc.subjectHistoryen_US
dc.subjectConvolutional Neural Networksen_US
dc.subjectDeep Priorsen_US
dc.subjectInverse Problemsen_US
dc.subjectPlug-And-Play Priorsen_US
dc.subjectSynthetic Aperture Radaren_US
dc.titlePlug-and-Play Synthetic Aperture Radar Image Formation Using Deep Priorsen_US
dc.typeArticle
dspace.entity.typePublication

Dosyalar

Koleksiyon

WoS İndeksli Yayınlar Koleksiyonu