Towards the use of deep generative models for the characterization in size of aggregated TiO2 nanoparticles measured by Scanning Electron Microscopy (SEM). (1st May 2019)

• Record Type: Journal Article
• Authors: Coquelin, L; Fischer, N; Feltin, N; Devoille, L; Felhi, G
• Is Part Of: Materials research express. Volume 6:Number 8(2019)
• Publication Date: 2019-05-01
• Subjects: SEM measurement -- generative adverserial networks -- deep learning -- particle size distribution; Materials science -- Research -- Periodicals; Materials science -- Periodicals; 620.11
• DOI: 10.1088/2053-1591/ab1bb4
• Languages: English
• ISSNs: 2053-1591

Abstract

Abstract: Recent advances in deep generative models based on convolutional neural networks (CNNs) are used to demonstrate the potential of these approaches for the estimation of particle size distribution on images of aggregated TiO2 particles obtained by Scanning Electron Microscopy (SEM). This very promising framework shall permit effective automation of SEM measurements analysis. Indeed, common image processing softwares bring the end-users with segmentation algorithms as well as measuring tools to estimate individual particle diameters. In the case of aggregated nanoparticles, most particles suffer missing contents and are not considered in the computations. In this paper, we use a recently developed method called 'context encoder's to predict missing parts of the nanoparticles. The approach is tested against simulated and real dropped image regions.