Classification of Mineral Phases in Unpolished Samples Using a Transformer Neural Network

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26220%2F26%3A0199984" target="_blank" >RIV/00216305:26220/26:0199984 - isvavai.cz</a>

Nalezeny alternativní kódy

RIV/00007064:K01__/25:N0000098

Výsledek na webu

<a href="http://dx.doi.org/10.1109/ICUMT67815.2025.11268630" target="_blank" >http://dx.doi.org/10.1109/ICUMT67815.2025.11268630</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/ICUMT67815.2025.11268630" target="_blank" >10.1109/ICUMT67815.2025.11268630</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Classification of Mineral Phases in Unpolished Samples Using a Transformer Neural Network

Popis výsledku v původním jazyce

Scanning electron microscopy (SEM) combined with energy dispersive spectroscopy (EDS) is widely used in geosciences for mineral phase classification. However, the lack of large, labeled datasets—especially for unpolished samples common in forensic pedology—limits the direct application of traditional supervised machine learning methods. In this study, we investigate the use of synthetic EDS data, generated using DTSA-II, to train advanced neural network architectures and evaluate their performance on real SEM–EDS measurements. We compare a U-Net baseline with a proposed 3D ResNet and a Transformer model. All models were trained on synthetic data and tested on real measurements of ten selected mineral phases. U-Net achieved accuracy of 63.7%, 3D ResNet reached 92.3%, and the Transformer model achieved the highest accuracy of 97.0%. These findings demonstrate that Transformer architectures can effectively generalize from synthetic to real EDS data, offering a promising way for accurate mineral phase classification in forensic and geological applications without the need for extensive labeled real datasets.

Název v anglickém jazyce

Classification of Mineral Phases in Unpolished Samples Using a Transformer Neural Network

Popis výsledku anglicky

Scanning electron microscopy (SEM) combined with energy dispersive spectroscopy (EDS) is widely used in geosciences for mineral phase classification. However, the lack of large, labeled datasets—especially for unpolished samples common in forensic pedology—limits the direct application of traditional supervised machine learning methods. In this study, we investigate the use of synthetic EDS data, generated using DTSA-II, to train advanced neural network architectures and evaluate their performance on real SEM–EDS measurements. We compare a U-Net baseline with a proposed 3D ResNet and a Transformer model. All models were trained on synthetic data and tested on real measurements of ten selected mineral phases. U-Net achieved accuracy of 63.7%, 3D ResNet reached 92.3%, and the Transformer model achieved the highest accuracy of 97.0%. These findings demonstrate that Transformer architectures can effectively generalize from synthetic to real EDS data, offering a promising way for accurate mineral phase classification in forensic and geological applications without the need for extensive labeled real datasets.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20201 - Electrical and electronic engineering

Projekt

<a href="/cs/project/VK01010107" target="_blank" >VK01010107: Aplikace umělé inteligence pro forenzní identifikaci stop zeminových fází</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Proceedings of the 17th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT)

ISBN

979-8-3315-7675-2

ISSN

—

e-ISSN

2157-023X

Počet stran výsledku

6

Strana od-do

170-175

Název nakladatele

—

Místo vydání

—

Místo konání akce

Florence, Italy

Datum konání akce

3. 11. 2025

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—