Automating electron microscopy: machine learning for cluster identification

点击翻译按钮获取中文摘要

Nanoparticles are of particular interest in the field of catalysis because of the high proportion of their atoms that are available at their surface (a high surface-to-volume ratio). Catalysis happens only at the surface of materials and is controlled by the electronic and spatial configuration of surface atoms. Nanoparticles of certain sizes are known to take up a small number of fixed shapes that possess well known configurations of atoms at their surface. Determining the shape of nanoparticles is a difficult question that requires very high-resolution characterisation techniques, such as electron microscopy.In this project, a convolutional neural network will be trained to recognize the different shapes of small nanoparticles. A convolutional neural network is a type of machine learning algorithm that can be trained to recognize image features. Once trained, the neural network will be used to determine the proportion of different particle shapes found in platinum nanoparticles of different sizes. This will determine which particle shapes have the lowest potential energy for each size and therefore will guide scientists to know which particles are likely to act as better catalysts for chemical reactions and processes.The trained neural network will be made available for anyone to use via its incorporation in to open-source software. This will allow anyone with electron microscope images of nanoparticles to use the same technique to analyse the shape of small nanoparticles.