본문 바로가기

Machine-learning-based semi-quantitative analysis model of clay minerals in Onsan-Busan coastal sediments, South Korea

문서 유형
연구출판물 · 학술지 논문
등재
KCISCIEScopus
저널 순위
JCR Q3Scopus Q2KCI Q3
저널
GEOSCIENCES JOURNAL30(3) · pp.400 · 2026
피인용
Web of Science 1Scopus 1KCI 0
FWCI (Scopus)
0
JCR IF
JCR 2024 1.5
ISSN
1226-4806

저자

  1. Park, Ju Young제1저자Kongju Natl Univ, Dept Geoenvironm Sci, Gongju Si 32588, Chungcheongnam, South Korea
  2. Jo, YesungKongju Natl Univ, Dept Geoenvironm Sci, Gongju Si 32588, Chungcheongnam, South Korea
  3. 구효진Pusan Natl Univ, Inst Future Earth, Busan 46241, South Korea
  4. Um, In KwonKorea Inst Geosci & Mineral Resources, Marine Geol Res Ctr, 124 Gwahak Ro, Daejeon 34132, South Korea
  5. Lee, Kyungbook교신저자Kongju Natl Univ, Dept Geoenvironm Sci, Gongju Si 32588, Chungcheongnam, South Korea

초록

Semi-quantitative analysis is performed for clay minerals based on the area ratios of individual peaks from X-ray diffraction (XRD), after over 16 h of organic matter pretreatment to separate intensity peaks. However, the analysis is time-intensive and subject to expert interpretation. This study developed machine-learning-based analysis models using the intensity profiles from air-dried or ethylene glycol-treated samples, Cases 1 and 2, respectively, from the Onsan-Busan coastal area. Additionally, Case 3 used only the main peaks selected by the Biscaye method. After developing a convolutional neural network-based model using 174 training and 75 validation samples, their performances were evaluated on 28 test samples. The output of the model is the compositional proportions of smectite, illite, kaolinite, and chlorite. Case 2 showed higher reliability, resulting in a 33% lower mean absolute error (MAE), compared to Case 1, because it used clearly separated smectite and chlorite peaks through pretreatment. Nevertheless, Case 1 maintained an acceptable MAE within 2 wt%, offering an efficient alternative without pretreatment. However, Case 3 showed superior performance, with an average MAE of 0.83 wt% and a coefficient of determination (R & sup2;) of 0.94, despite using approximately 30% of the original 1,337 input values, or 492 intensities, by focusing on the main peaks. Furthermore, applying Case 3 to 12 unseen samples from the Pohang area with transfer learning yielded an MAE of 3.28 wt% and R & sup2; of 0.70, confirming the scalability of the model to new regions. In conclusion, Case 3 provides useful guidance for expert analysis, enhancing analytical efficiency.

키워드

Machine-learningX-ray diffraction (XRD)Clay mineralSemi-quantitative analysisFeature selection

식별자

Scopus EID
2-s2.0-105034299782
WoS UID
001720969200001
KCI
ART003337402