Example data set for Jupyter notebook

High-resolution transmission electron microscopy (HRTEM) is an important technique for investigating nanoparticles at atomic resolution. One drawback is that the intense electron beam required for sufficient electron signal can be harmful for the sample. Lowering the electron beam intensity (electron dose rate) can cause less damage to the sample. However, the latter can result in images drowning in noise. Novel techniques applying neural networks in machine learning can be applied to detect events during a series of HRTEM images recorded at low electron dose rate.

To assist the machine learning approach, novel signal-to-noise ratio (SNR) models are applied to series of HRTEM images at varied electron dose rates. Furthermore, a novel approach of structural similarity index measurement (SSIM), where each frame is compared to an adjacent frame, is applied to the HRTEM image series.

The HRTEM image series consist of .dm4 files for each point in time. All files are compressed into a folder (.zip). The dataset provided is an example of 50 frames, each recorded at 0.2 s exposure time, at a fixed dose rate.

Obtaining SNR and SSIM values of the HRTEM image series is done using Jupyter notebooks. An example of such is obtainable at a GitLab repository: https://gitlab.com/wibang_dtu_91dk/doserate-snr-ssim/

The Jupyther notebook loads the individual .dm4 files into a stack. Using a specific package called HyperSpy, the user can browse through the series and subsequently extract data from selected areas used for the SNR and SSIM. The Jupyter notebook exports the data into data sheets (.csv), which can be loaded into other scripts for further treatments. Finally, the script can also export the data as a video (.mp4 and .gif) at a specified frame rate.

The main work referring to the Jupyter notebook and dataset is a recently accepted article: W. B. Lomholdt, M. H. L. Larsen, C. N. Valencia, J. Schiøtz and T. W. Hansen, "Interpretability of high-resolution transmission electron microscopy images", Ultramicroscopy, vol. 263, 2024, pp. 113997, https://doi.org/10.1016/j.ultramic.2024.113997