|About this Abstract
||2019 TMS Annual Meeting & Exhibition
||Computational Approaches for Big Data, Artificial Intelligence and Uncertainty Quantification in Computational Materials Science
||Unsupervised Segmentation of Microstructures
||Bo Lei, Elizabeth Holm
|On-Site Speaker (Planned)
Microstructure analysis is crucial to materials science and engineering. The central part of microstructure analysis is the segmentation of different phases, grains or regions based on their visual difference in microscopic images. To automatically perform the segmentation task, people have built specialized image processing pipelines. However, the traditional methods are often limited to a particular microstructure system. Recently, computer vision and machine learning methods are being applied to image analysis with great advantage in performance and generality. The supervised methods require experts to annotate many images pixel-wise, which takes a great effort for large and high-resolution datasets. We have developed an unsupervised method based on deep learning and clustering without the need for annotation. The unsupervised method achieves a robust result on the Ultrahigh Carbon Steel (UHCS) microstructure dataset.
||Planned: Supplemental Proceedings volume