Analysis of TEM data using machine learning methods

Abstract

Liquid phase electron microscopy has various advantages over other in situ microscopy techniques. Due to the nature of the liquid medium during sampling, various types of specimens can be observed, which would not have been suitable in a typical vacuum setup. An additional advantage would be the ability to observe samples without the need for traditional sample preparation. In particular, there is a degree of inaccuracy in the observed morphology of polymer particles when placed under vacuum conditions, without an liquid medium. As such, a liquid medium would directly address this issue and there would be a greater degree of accuracy in its observed morphology. However, liquid phase electron microscopy does come with limitations, which poses a problem for clear and accurate imaging of the samples. These imaging limitations are further exacerbated by the need for low electron doses to preserve the sample. Therefore this paper shall discuss the limitations of liquid phase imaging and the effect of low electron dosage. This paper also presents a comparison of imaging techniques, such as filtering and machine learning methods, that would improve the quality of the raw liquid phase images.