Online Lecture by prof. Lionel Germain

Deep Learning applied to microstructure analysis. Performances and limitations.

Laboratoire d’Étude des Microstructures et de Mécanique des Matériaux
Université de Lorraine – Site Technopole, Metz, France

Title:
Deep Learning applied to microstructure analysis. Performances and limitations.

Abstract:
Deep learning has revolutionized image processing in many applications (computer vision, image generation …). Applied to microstructure analyses, it has also shown very good performances in many different tasks. In particular, convolutional neural networks (CNNs) have enabled significant advancements in tasks like classification, segmentation, and feature extraction from microstructural images. Often, these methods have demonstrated state of the art performances (when not superior) in identifying and quantifying microstructural features compared to traditional techniques.

In this presentation, the working principles of CNNs will be shown. Several examples of applications will be presented and their performances and limitation will be discussed. The examples addressed both optical microstructures, SEM micrographs and EBSD data. The specificity of each example will be discussed.

This session will be chaired by prof. Leo Kestens (UGent – TU Delft)