Scanning electron microscopy (SEM) is a powerful noninvasive method of examination of microscopic structures, which has spatial resolution 1-100 nm. The principle of SEM operation is based on irradiation of the object under study with a monoenergetic electron beam and subsequent registration of intensity and/or energy distribution of electrons in the beam after reflection by the object (reflected SEM). The problem of microtomography in reflected electrons is a classical ill-posed problem of mathematical physics. Standard methods of reflected electron tomography usually only aim at separating superimposed images of layers rather than investigating their characteristics or composition. The current project involves reconstructing the internal structure of a layered object (cladding) in terms of layer thickness, material density. In 2021 convolutional neural networks were effectively used to solve the inverse problem of fluorescence microtomography. CNN networks (UNet, 3-d UNer) presumably can be used as effective solver of reflected electron microtomography inverse problem. The project is aimed to test and compare different CNN-architectures on volumetric layered structure.
