IT-16-P-3032 Computer vision in the service of Crystallography: Automated analysis of atomic-resolution images

An automated tool for a crystallographic analysis of HRTEM (High resolution Transmission electron microscopy), HRSTEM (High resolution Scanning Transmission electron microscopy) and diffraction images is proposed. Algorithms of artificial intelligence and computer vision are employed to detect features carrying the information and to process them in order to determine or estimate crystallographic quantities. This shall result in an expedited analysis, possibly higher precision and little to no human effort compared to manual analysis.

In the case of SAD (Selected area diffraction) images, diffraction spots or disks are detected in the widest possible area of the pattern and the zone axis is calculated. If the observed material is not known, the tool can choose the most probable candidate from a list candidates. HREM (High resolution electron microscopy) images can be segmented to separate individual grains depicted. If the image contrast features directly correspond to positions of atomic columns, the zone axis is determined and the crystallographic planes and direction in the image are identified. Dislocation detection and quantification can be performed as well as a grain misorientation estimation, reconstruction of positions of individual atoms and so on. If the image contrast features do not correspond directly to the atomic column positions, the atomic columns can be found using HREM simulations.

The proposed tool (implemented in MATLAB) has been successfully tested on number of real world images and diffraction patterns. It has proven its ability to autonomously provide correct results.