主催: 一般社団法人 日本機械学会
会議名: 第25回 動力・エネルギー技術シンポジウム
開催日: 2021/07/26 - 2021/07/27
A framework for automated SOFC microstructure reconstruction from large asymmetric-resolution FIB-SEM datasets is proposed. Machine learning techniques are used for super-resolving the in-depth direction, i.e. FIB slicing direction, and for automating the phase segmentation. Deep neural networks consisting of patch-VDSR residual network for the increasing slicing resolution of FIB-SEM data and patch-CNN in the encoder-decoder configuration for semantic segmentation are incorporated. The proposed algorithm can shorten the FIB-SEM measurement time or increase the size of microstructures maintaining high resolution.