Bulat Bakirov , Sergey E. Kichanov , Denis P. Kozlenko
{"title":"Convolutional neural networks for reconstruction of neutron tomography from incomplete data","authors":"Bulat Bakirov , Sergey E. Kichanov , Denis P. Kozlenko","doi":"10.1016/j.nimb.2025.165682","DOIUrl":null,"url":null,"abstract":"<div><div>To reduce the number of image projections in neutron tomography experiments, tomography reconstruction algorithms are used from an incomplete and limited number of neutron projections. The possibilities of a reconstruction algorithm based on convolutional neural networks are presented. It was found that only 72 projections are required for trained convolutional neural network for a qualitative reconstruction comparable to that from a complete dataset. Variations in the quality of the tomography reconstruction due to changes in training data and the number of input projections are considered. Examples of using convolutional neural networks for neutron tomography reconstructions of real experimental data includes the results of studies on archaeological metal materials from the “Volna-1″ archaeological site.</div></div>","PeriodicalId":19380,"journal":{"name":"Nuclear Instruments & Methods in Physics Research Section B-beam Interactions With Materials and Atoms","volume":"563 ","pages":"Article 165682"},"PeriodicalIF":1.4000,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Nuclear Instruments & Methods in Physics Research Section B-beam Interactions With Materials and Atoms","FirstCategoryId":"101","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0168583X25000722","RegionNum":3,"RegionCategory":"物理与天体物理","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"INSTRUMENTS & INSTRUMENTATION","Score":null,"Total":0}
引用次数: 0
Abstract
To reduce the number of image projections in neutron tomography experiments, tomography reconstruction algorithms are used from an incomplete and limited number of neutron projections. The possibilities of a reconstruction algorithm based on convolutional neural networks are presented. It was found that only 72 projections are required for trained convolutional neural network for a qualitative reconstruction comparable to that from a complete dataset. Variations in the quality of the tomography reconstruction due to changes in training data and the number of input projections are considered. Examples of using convolutional neural networks for neutron tomography reconstructions of real experimental data includes the results of studies on archaeological metal materials from the “Volna-1″ archaeological site.
期刊介绍:
Section B of Nuclear Instruments and Methods in Physics Research covers all aspects of the interaction of energetic beams with atoms, molecules and aggregate forms of matter. This includes ion beam analysis and ion beam modification of materials as well as basic data of importance for these studies. Topics of general interest include: atomic collisions in solids, particle channelling, all aspects of collision cascades, the modification of materials by energetic beams, ion implantation, irradiation - induced changes in materials, the physics and chemistry of beam interactions and the analysis of materials by all forms of energetic radiation. Modification by ion, laser and electron beams for the study of electronic materials, metals, ceramics, insulators, polymers and other important and new materials systems are included. Related studies, such as the application of ion beam analysis to biological, archaeological and geological samples as well as applications to solve problems in planetary science are also welcome. Energetic beams of interest include atomic and molecular ions, neutrons, positrons and muons, plasmas directed at surfaces, electron and photon beams, including laser treated surfaces and studies of solids by photon radiation from rotating anodes, synchrotrons, etc. In addition, the interaction between various forms of radiation and radiation-induced deposition processes are relevant.