A Method for High-Throughput Processing of STEM Datasets.

The quest for efficient exploration and characterisation of novel materials has prompted the development of advanced methods in synthesising and analysing nanomaterials. In this context, our research focuses on the development of routines towards automated analysis of (scanning) transmission electron microscopy ((S)TEM) datasets. To complement this, we propose the integration of a machine learning-assisted algorithm for the identification of individual nanocrystals in high annular dark field images (HAADF) obtained through STEM. This algorithm is essential for acquiring statistically relevant data on the chemical composition of individual heterogeneous nanocrystals, particularly those generating minimal X-rays due to their limited thickness or beam sensitivity. The proposed algorithm aims to address the challenges associated with identifying nanocrystals in TEM images, where the HAADF signal is linearly proportional to thickness and approximately to the average atomic number squared (Z²). This information is crucial for understanding the characteristics of nanocrystals, including lateral size and signal intensity. The need for automation arises from the limitations of existing methods, emphasising the significance of a streamlined and objective approach to data analysis. The approaches developed in the analysis of nanoparticles ensembles were then extended to the treatment of datasets to reconstruct the 3D chemical distribution in nanoparticles assemblages and the denoising of spectroscopic data from cathodoluminescence measurements.