61 XAFS-CT (X-ray Absorption Fine Structure Computed Tomography) is a technique that measures the internal structure and chemical states of materials by varying the projection direction and energy of X-rays. Faster measurements using a limited number of projections, namely sparse-view CT, are necessary. This can be addressed by assuming that CT images are sparse, a principle known as compressed sensing [1]. In this study, we investigated the applicability of sparse-view CT for the reconstruction of multiple coefficient data in XAFS-CT, targeting MOF (Metal-Organic Framework) materials, which are expected to have societal applications as adsorbents [2]. As a result, the compressed sensing method (FISTA) achieved higher reconstruction accuracy with fewer projections than the conventional OSEM (Ordered Subset Expectation Maximization) (Figure 1). Poster Award NomineePoster Award NomineeHigh-speed measurement of MOF using sparse-view XAFS-CT reconstruction with compressed sensingP2-09Sparse Coding-Based Multiframe Superresolution for Efficient Synchrotron Radiation Microspectroscopy Yasuhiko Igarashi1,2, Naoka Nagamura2,3,4, Masahiro Sekine1, Hirokazu Fukidome4, Hideitsu Hino5,and Masato Okada2,6 1 University of Tsukuba, 2 Photoemission Group, National Institute for Materials Science, 3 Tokyo University of Science, 4 Tohoku University, 5 The Institute of Statistical Mathematics, 6 The University of Tokyo In nanostructure extraction, advanced techniques like synchrotron radiation and electron microscopy are often hindered by radiation damage and charging artifacts from long exposure times. This study presents a multiframe superresolution method using sparse coding to enhance synchrotron radiation microspectroscopy images. By reconstructing high-resolution images from multiple low-resolution ones, exposure time is minimized, reducing radiation effects, thermal drift, and sample degradation while preserving spatial resolution. Unlike deep learning-based superresolution methods, which overlook positional misalignment, our approach treats positional shifts as known control parameters, enhancing superresolution accuracy with a small, noisy dataset. Unlike state-of-the-art deep learning techniques that require large datasets, our method excels with limited data, making it ideal for real-world scenarios with constrained sample sizes. This approach offers enhanced image quality, reduced exposure times, and improved interpretability of scientific data, making it a versatile tool for overcoming the challenges associated with radiation damage and sample degradation in nanoscale imaging. P2-10Naoki Yamane1, Hirosuke Matsui2, Mizuki Tada2, and Yasuhiko Igarashi3,41 Graduate School of Science and Technology, University of Tsukuba2 Graduate School of Science, Nagoya University3 Institute of Systems and Information Engineering, University of Tsukuba4 Photoemission Group, National Institute for Materials Science (NIMS)[1] H.Kudo et al., QIMS., 3, 14761–14161 (2013).[2] E.Yamada et al., J. Am. Chem. Soc., 146, 9181–9190(2024).
元のページ ../index.html#61