Japan Geoscience Union Meeting 2025

Presentation information

[J] Oral

M (Multidisciplinary and Interdisciplinary) » M-IS Intersection

[M-IS14] Paleoclimatology and paleoceanography

Fri. May 30, 2025 1:45 PM - 3:15 PM Convention Hall (CH-A) (International Conference Hall, Makuhari Messe)

convener:Takashi Obase(Japan Agency for Marine-Earth Science and Technology), Atsuko Yamazaki(Graduate School of Environmental Studies, Nagoya University), Hitoshi Hasegawa(Faculty of Science and Technology, Kochi University), Yusuke Okazaki(Department of Earth and Planetary Sciences, Graduate School of Science, Kyushu University), Chairperson:Hitoshi Hasegawa(Faculty of Science and Technology, Kochi University)


1:45 PM - 2:00 PM

[MIS14-21] Automated classification of radiolarians in the Northwest Pacific: Advancing high-resolution analysis of fossil assemblages using a Virtual Slide Scanner and Deep Learning

*Takuya Itaki1, Isao Motoyama2, Kenji Matsuzaki4, Shin-ichi Kamikuri3, Kazuhide Mimura1, Ayumu Miyakawa1 (1.Geological Survey of Japan, AIST, 2.Yamagata University, 3.Ibaraki University, 4.University of Tokyo)

Keywords:microfossil, faunal analysis, automation, high throughput

Radiolarians are important indicator organisms in paleoceanographic analysis, and their classification and analysis play a crucial role in geological research. The aim of this study is to develop a method for the automated classification of radiolarians in the Northwest Pacific Ocean by using a virtual slide scanner to obtain image data and training a deep learning model. A virtual slide scanner is a device that captures and observes microscope slides as digital image data (virtual slides). The system operated by Geological Survey of Japan uses image analysis software to perform focus synthesis and particle image extraction from virtual slide data, and automatically classifies particle images using deep learning. In this study, a total of 50,000 images obtained with the virtual slide scanner NanoZoomer S-360 (Hamamatsu Photonics K.K.) were prepared as training data, and the classification accuracy of 30 species of radiolarians, which are useful as paleoceanographic indicators, is currently being investigated. Preliminary experiments using 271 layers of a sediment core collected off the coast of Boso Peninsula have yielded high-resolution records of the cold intermediate water species Cycladophora davisiana and the warm Kuroshio species Tetrapyle spp. over the past 15,000 years, with results consistent with previous reports. This method is expected to enable more efficient and accurate classification compared to traditional manual classification, opening up new possibilities for the analysis of fossil assemblages.