"Deep Learning based segmentation and tracking of nanoparticles on surfaces applied to the in situ study of nanocatalysts by Scanning Transmission Electron Microscopy" by Christophe Ducottet

Seminar by Christophe Ducottet, Laboratoire Hubert Curien

Abstract

The context of this study is to better control the conditioning of nanocatalysts by monitoring the evolution and movement of metallic nanocatalyst particles on an oxide support during in situ experiments under gas and temperature in an Environmental TEM. From a sequence of images acquired in the STEM mode, the objective is thus to recover the trajectory of all the nanoparticles (NPs) visible on the support in order to quantify their evolution in time (morphology, number, events such as coalescence or disappearance). In this presentation, we will focus on the two main steps of our approach: the segmentation and the tracking of NPs. For segmentation, we propose to use the well known UNet deep learning architecture with a specific training procedure relying on realistic simulated images. Then, the multi-target tracking is formulated as a specific energy minimization which takes into account physical constraints on the trajectories and consider discrete jump moves to change the dimensionality of the model.

This seminar will be held in French\English

Cisco Webex : https://ujmstetienne.webex.com/meet/ImageSCV