Presentation Information
[R1P-03]Identification of rock type from thin sections using CNN deep learning<font color="#FF7F27">「発表賞エントリー」</font>
*Mana Matsuda1, Yusuke Seto1 (1. Kobe Univ. Sci.)
Keywords:
deep learning,rock classification,computer vision
In the research, we demonstrated a classification of various rock types by deep learning from petrographic thin sections. In order to acquire a large number of training data, a high-speed automatic shooting system was incorporated into a polarizing microscope. 7200 images of petrographic thin sections (ten rock types) were used for deep learning. As a result, using a stack image of open/crossed Nicol images as input format, the accuracy rate of the classification test reached 98%, suggesting that the deep learning method is an effective tool for identification of rock types.