Nanotalk - How can deep learning improve the quantitative treatment of in situ TEM experiments?

In this DENSsolutions Wildfire Nanotalk, CNRS Research Director at Ircelyon, University of Lyon, Dr. Thierry Epicier, presents his research on the application of machine/deep learning approaches to in situ TEM experiments.

Simply enter your email address above and click "Watch now" to register. You will receive an email directly after with a link to watch this Nanotalk.

Abstract

The availability of instrumentalized specimen holders, environmental montages and fast recording cameras has contributed to the development of in situ approaches in Transmission Electron Microscopy (TEM) studies in all fields of materials and life sciences. Therefore, huge series of data are acquired, such as time-lapse, video-type sequences of images. This raises two major issues for scientists: how can these results be treated in a reasonable time frame, and how quantitative and statistically representative can these treatments be? In this context, applying machine/deep learning approaches can drastically help to provide a better, faster and more objective analysis of data as compared to a traditional, subjective and tedious manual counting and interpretation.

During this Nanotalk, Dr. Thierry Epicier illustrates the application of a combined deep learning and computer vision approach to track the evolution of a population of supported nanoparticles during in situ ETEM experiments under gas and temperature conditions using our Wildfire system. The experiments he presents involve Pd(O) nanoparticles supported on delta-alumina and Pt nanoparticles on gamma-alumina at high temperatures under oxygen or hydrogen gaseous fluxes.

Dr. Epicier first demonstrates that the U-net convolutional neural network allows to determine the position, size of large amounts of nanometric particles as a function of time during heating experiments. Next, he shows that you can combine these measurements with a computer vision approach based on a multiple-objects tracking algorithm. A quantitative analysis of the nanoparticle trajectories can then be performed, allowing you to detect interaction events such as the fusion (coalescence) or disappearance of individual objects.

Register now for free to watch this Nanotalk!