Machine learning for the classification of serial electron diffraction patterns: synthetic data

Serial electron crystallography faces a fundamental challenge due to the flat Ewald sphere resulting from the short electron wavelength, leading to limited 3D information in individual patterns. Recently, an algorithm for unit-cell determination from zonal electron diffraction patterns (GM algorithm) [Miehe (1997). Ber. Dtsch. Miner. Ges. Beih. z. Eur. J. Miner. 9, 250; Gorelik et al. (2025). Acta Cryst. A81, 124-136] was introduced in the context of serial electron crystallography. This algorithm requires the extraction of 2D zonal patterns from the complete serial dataset. Here, we present a machine learning approach for pattern sorting and apply it initially to simulated electron diffraction patterns.