Big data analysis techniques in X-ray fluorescence imaging spectroscopy
Φόρτωση...
Ημερομηνία
Συγγραφείς
Γεροδήμος, Θεοφάνης
Τίτλος Εφημερίδας
Περιοδικό ISSN
Τίτλος τόμου
Εκδότης
Πανεπιστήμιο Ιωαννίνων. Πολυτεχνική Σχολή. Τμήμα Μηχανικών Επιστήμης Υλικών
Περίληψη
Τύπος
Είδος δημοσίευσης σε συνέδριο
Είδος περιοδικού
Είδος εκπαιδευτικού υλικού
Όνομα συνεδρίου
Όνομα περιοδικού
Όνομα βιβλίου
Σειρά βιβλίου
Έκδοση βιβλίου
Συμπληρωματικός/δευτερεύων τίτλος
Τεχνικές ανάλυσης μεγάλων δεδομένων στην απεικονιστική φασματοσκοπία φθορισμού ακτίνων-Χ
Περιγραφή
Big data presents exciting opportunities and huge challenges for data scientists. That is why
this field is very promising and considered cutting-edge in many scientific disciplines. The ability
to perform complex calculations and extract information from large datasets is a valuable research
tool in both basic and applied research. The size and complexity (high dimensionality) of big data
present unique challenges, including scalability and storage limitations, noise over clustering,
misleading correlations, and randomness errors or problems during measurements. All these
challenges require innovative computational and statistical approaches.
X-ray fluorescence spectroscopy (XRF) is an analytical method widely used in many fields
due to its nondestructive and noninvasive nature. XRF spectroscopy allows multi-elemental
determination, requires no processing of the sample, provides fast results, and is environmentally
friendly. XRF is based on detecting the specific radiative transitions emitted by the atoms of a
material when it is exposed to a primary X-ray beam.
The rapid technological development in the last two decades in the field of X-ray tubes,
optical devices, as well as energy dispersive detectors has resulted in the rapid development of Xray fluorescence imaging spectroscopy. In X-ray fluorescence imaging spectroscopy, the ionizing
beam scans the target, allowing the elemental composition of the target to be determined based
on its position coordinates. The elemental composition is determined from the spectrum recorded
in each spatial area during scanning. The number of spectra during a scan depends on the beam
spot, the pixel size, and the size of the object, and can exceed a million. Furthermore, each
spectrum contains information in thousands of channels.
Consequently, the analysis of big data produced by XRF imaging spectroscopy is crucial due
to the complexity and volume of data generated during the process. The techniques for analyzing
the experimental data can improve the strength of this analytical method. The XRF spectra are
high-dimensional, containing vast amounts of data points (spectra) for each sample analyzed. This
complexity arises from the need to detect and measure the intensity of emitted X-rays across a
broad spectrum of energies. Advanced analytical techniques, such as Principal Component
Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Non-negative Matrix
Factorization (NMF), are essential for dimensionality reduction and data simplification. These
techniques enable researchers to manage and interpret the high-dimensional data effectively, extracting meaningful patterns and correlations that would be otherwise hidden/unrevealed.
The application of clustering algorithms like k-means and machine learning models,
including neural networks, further enhances the capability to process and analyze large XRF
datasets. These models facilitate the identification of distinct elements within samples and
improve the accuracy of qualitative and quantitative analysis. Moreover, integrating machine deep
learning architectures like convolutional neural networks (CNNs) and auto encoders, expands the
analytical power of XRF spectroscopy. These algorithms enable the automation of data
processing, reducing the need for manual intervention and special knowledge and increasing the
efficiency of the analytical workflow.
Chapters 1 and 2 introduce the basics of X-rays and X-ray fluorescence (XRF) spectroscopy.
They cover the main theory of X-rays, the instrumentation of XRF spectrometers, and the
challenges associated with analyzing and interpreting the complex data generated.
In Chapters 3 and 4 advanced analytical techniques, such as dimensionality reduction and
machine learning algorithms, are discussed that help researchers extract meaningful information
from this big data. Subsequent chapters delve deeper into the application of these advanced
analytical methods in specific case studies, illustrating the practical benefits of big data analysis
in XRF spectroscopy.
In Chapter 5, we demonstrate the effectiveness of big data analysis techniques collected
during the application of macroscopic scanning X-ray fluorescence spectroscopy to a Byzantine
religious icon (Macroscopic XRF, MA-XRF). By comparing methods of X-ray fundamental
parameter analysis and statistical data analysis, we demonstrate how these approaches can extract
detailed information about pigments used, painting techniques, and the state of preservation of
historical artifacts.
In Chapter 6, the fusion of datasets acquired from the same object applying asynchronously
scanning micro-X-ray fluorescence and multispectral imaging spectroscopy (MSI) is explored.
Specifically, the study of stamps was chosen, due to the increased requirements of spatial
resolution. This chapter highlights the enhanced analytical capabilities achieved through data
fusion, emphasizing the importance of aligning and co-registering datasets from different imaging
techniques to obtain more comprehensive insights.
Chapter 7 examines the application of Artificial Intelligence (AI) methods to the analysis of
datasets acquired by applying MA-XRF on a 19th-century religious image. This chapter illustrates how clustering algorithms, factorization methods, and supervised machine learning techniques
can provide detailed and rapid analysis, making advanced data interpretation accessible even to
non-experts.
Chapter 8 focuses on using convolutional neural networks (CNNs) to analyze MA-XRF
datasets from religious panel paintings. It highlights CNNs' effectiveness in accurately identifying
and mapping elemental transitions, thus enabling automated spectral analysis and providing
support for novice and experienced analysts.
Chapter 9 presents two case studies of machine learning applied to art conservation analysis.
Firstly, investigates the potential of CNN classifiers to analyze complex multilayer samples in
paintings. By training CNNs to predict paint layers from scanning XRF spectra, this chapter
demonstrates the capability of these networks to reveal the stratigraphy of artworks, offering
valuable tools for art conservation and study. Secondly, investigates the per-pixel correlation
between RGB images and XRF spectra using a deep autoencoder, providing a comprehensive
understanding of the relationship between visual and spectral data in artworks before restoration
and conservation.
In conclusion, big data analysis in XRF spectroscopy is a crucial tool for analyzing the large
datasets acquired during the measurements. By leveraging advanced data analysis techniques and
machine learning algorithms, researchers can more effectively handle the complexity and volume
of XRF data, leading to more accurate and reliable results. This synergy between domains like
XRF spectroscopy and big data analysis ultimately drives advancements in material science,
quality control, cultural heritage studies, and various industrial applications.
Περιγραφή
Λέξεις-κλειδιά
Φασματοσκοπία εκπομπής ακτίνων-Χ, Ανάλυση μεγάλων δεδομένων, Μηχανική Μάθηση
Θεματική κατηγορία
Μεγάλα δεδομένα -- Ανάλυση
Παραπομπή
Σύνδεσμος
Γλώσσα
en
Εκδίδον τμήμα/τομέας
Πανεπιστήμιο Ιωαννίνων. Πολυτεχνική Σχολή. Τμήμα Μηχανικών Επιστήμης Υλικών
Όνομα επιβλέποντος
Αναγνωστόπουλος, Δημήτριος
Εξεταστική επιτροπή
Αναγνωστόπουλος, Δημήτριος
Λύκας, Αριστείδης
Καρύδας, Ανδρέας Γερμανός
Ματίκας, Θεόδωρος
Παπαγεωργίου, Δημήτριος
Μαστροθεόδωρος, Γεώργιος
Παπαδάκης, Βασίλειος
Λύκας, Αριστείδης
Καρύδας, Ανδρέας Γερμανός
Ματίκας, Θεόδωρος
Παπαγεωργίου, Δημήτριος
Μαστροθεόδωρος, Γεώργιος
Παπαδάκης, Βασίλειος
Γενική Περιγραφή / Σχόλια
Ίδρυμα και Σχολή/Τμήμα του υποβάλλοντος
Πανεπιστήμιο Ιωαννίνων. Πολυτεχνική Σχολή. Μηχανικών Επιστήμης Υλικών
Πίνακας περιεχομένων
Χορηγός
Βιβλιογραφική αναφορά
Ονόματα συντελεστών
Αριθμός σελίδων
249
Λεπτομέρειες μαθήματος
Συλλογές
item.page.endorsement
item.page.review
item.page.supplemented
item.page.referenced
Άδεια Creative Commons
Άδεια χρήσης της εγγραφής: Attribution-NonCommercial-NoDerivs 3.0 United States