TY - JOUR
T1 - Visualization of judgment regions in convolutional neural networks for X-ray diffraction and scattering images of aliphatic polyesters
AU - Amamoto, Yoshifumi
AU - Kikutake, Hiroteru
AU - Kojio, Ken
AU - Takahara, Atsushi
AU - Terayama, Kei
N1 - Funding Information:
Acknowledgements The synchrotron radiation experiments were performed at the BL40B2 and BL40XU beamlines of SPring-8 with the approval of the Japan Synchrotron Radiation Research Institute (JASRI) (Proposal no. 2020A1525 and 2019B1667). This work was supported by the Cabinet Office, Government of Japan, Cross-ministerial Strategic Innovation Promotion Program (SIP), “Technologies for Smart Bio-industry and Agriculture” (funding agency: Bio-oriented Technology Research Advancement Institution, NARO). The computer resources were supported by the “Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures” and the “High Performance Computing Infrastructure” in Japan (Project ID: jh200016-NAH). This work was also supported by the JSPS Grant-in-Aid for Scientific Research on Innovative Areas, Discrete Geometric Analysis for Materials Design: 20H04644, by the Grant-in-Aid for Scientific Research (B): 20H02800, and by Early-Career Scientists: 18K14273 from JSPS. YA and KT acknowledge the financial support of the Grant-in-Aid for RIKEN-Kyushu University Science and Technology Hub Collaborative Research Program.
Publisher Copyright:
© 2021, The Author(s), under exclusive licence to The Society of Polymer Science, Japan.
PY - 2021/11
Y1 - 2021/11
N2 - The construction of a deep learning model and visualization of judgment regions were conducted for X-ray diffraction and scattering images of aliphatic polyesters. Due to recent progress in measurement methods, a large amount of image data can be obtained in a short time; therefore, machine learning methods are useful to determine the important regions for a given objective. Although techniques to visualize the judgment regions using deep learning have recently been developed, there have been few reports discussing whether such models can determine the important regions of X-ray diffraction and scattering images of polymeric materials. Herein, we demonstrate classification models based on convolutional neural networks (CNNs) for wide-angle X-ray diffraction and small-angle X-ray scattering images of aliphatic polyesters to predict the types of polymers and several crystallization temperatures. Furthermore, the judgment regions of the X-ray images used by the CNNs were visualized using the Grad-CAM, LIME, and SHAP methods. The main regions were diffraction and scattering peaks recognized by experts. Other areas, such as the beam centers were recognized when the intensity of the images was randomly changed. This result may contribute to developing important features in deep learning models, such as the recognition of structure–property relationships.
AB - The construction of a deep learning model and visualization of judgment regions were conducted for X-ray diffraction and scattering images of aliphatic polyesters. Due to recent progress in measurement methods, a large amount of image data can be obtained in a short time; therefore, machine learning methods are useful to determine the important regions for a given objective. Although techniques to visualize the judgment regions using deep learning have recently been developed, there have been few reports discussing whether such models can determine the important regions of X-ray diffraction and scattering images of polymeric materials. Herein, we demonstrate classification models based on convolutional neural networks (CNNs) for wide-angle X-ray diffraction and small-angle X-ray scattering images of aliphatic polyesters to predict the types of polymers and several crystallization temperatures. Furthermore, the judgment regions of the X-ray images used by the CNNs were visualized using the Grad-CAM, LIME, and SHAP methods. The main regions were diffraction and scattering peaks recognized by experts. Other areas, such as the beam centers were recognized when the intensity of the images was randomly changed. This result may contribute to developing important features in deep learning models, such as the recognition of structure–property relationships.
UR - http://www.scopus.com/inward/record.url?scp=85111664936&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85111664936&partnerID=8YFLogxK
U2 - 10.1038/s41428-021-00531-w
DO - 10.1038/s41428-021-00531-w
M3 - Article
AN - SCOPUS:85111664936
VL - 53
SP - 1269
EP - 1279
JO - Polymer Journal
JF - Polymer Journal
SN - 0032-3896
IS - 11
ER -