planssetr.blogg.se

We have also built a TEM ImageNet project website for searching, browsing, and downloading of the training images and labels 13. The app is free and open-source and it is available for download on Github 12. Taking a step further, we have deployed our models to a desktop app with a graphical user interface. Herein, we report the development of a training library and a deep learning method that can perform robust and precise atom segmentation, localization, denoising, and deblurring/super-resolution processing of experimental images.

For instance, without human supervision, it is non-trivial to localize the dimmer atomic columns on or near the edge/surface of a particle due to the lower contrast and intensities. Particularly, for atomic-scale scanning TEM images, to date there is no established algorithm that is sufficiently robust to detect all atomic features when there is large thickness change in an image. Although several algorithms including graph methods 4, clustering methods 5, 6, 7, 8, 9, threshold methods 10 and edge detection methods 11 can achieve reasonable performance in pre-defined sceneries, they tend to fall short when noises are strong and interferences are unpredictable. It will not only be a valuable tool to student researchers and materials scientists who use ADF-STEM as a tool but also can assist experienced electron microscopists in automated analysis of large datasets.Ītomic column localization and segmentation in atomic-resolution scanning TEM images with high precision and high robustness is non-trivial. Such methods, if available, can greatly reduce misinterpretation, bias, and human errors.

Therefore, it is highly desirable to develop a robust method to detect and localize atoms/atomic columns and restore the atomic-scale information in non-ideal ADF-STEM images.

In non-ideal ADF-STEM images that are contaminated by noise and distortions, the atomic arrangement might still be recognizable by experienced electron microscopists, but some low-contrast atomic details might not be easily detectable by inexperienced operators. However, acquiring and maintaining these high-resolution instruments incur high costs and to date recording high quality atomic-scale data is still a time-consuming process-high-quality STEM images are not always available, due to many environmental factors, such as scan jittering, temperature fluctuations, stray electromagnetic fields, sample charging and drifting. Recently, Muller and his coauthors have demonstrated that by combining a ptychography technique with a highly sensitive pixelated detector, the resolution envelope can be extended to 39 pm even at low acceleration voltages (80 keV), a condition that can greatly reduce electron beam damage to low-atomic-number materials while retaining ultrahigh resolution 3. Using advanced aberration-corrected ADF-STEM, direct acquisition of real-space images with 50-pm resolution can be achieved at high acceleration voltages (300 keV) 1, 2. In the past decade, the widespread availability of aberration-corrected annular dark-field scanning transmission electron microscopy (ADF-STEM) that offers reliable atomic-scale imaging of materials, has enormously benefited many fields ranging from nanocatalysts and batteries to electronic and structural materials. We have also built a TEM ImageNet project website for easy browsing and downloading of the training data. Taking a step further, we have deployed our deep-learning models to a desktop app with a graphical user interface and the app is free and open-source. Despite using simulated images as training datasets, the deep-learning model can self-adapt to experimental STEM images and shows outstanding performance in atom detection and localization in challenging contrast conditions and the precision consistently outperforms the state-of-the-art two-dimensional Gaussian fit method. Herein, we report the development of a training library and a deep learning method that can perform robust and precise atom segmentation, localization, denoising, and super-resolution processing of experimental images. Particularly, for atomic-resolution STEM images, so far there is no well-established algorithm that is robust enough to segment or detect all atomic columns when there is large thickness variation in a recorded image. Although several conventional algorithms, such has thresholding, edge detection and clustering, can achieve reasonable performance in some predefined sceneries, they tend to fail when interferences from the background are strong and unpredictable. Atom segmentation and localization, noise reduction and deblurring of atomic-resolution scanning transmission electron microscopy (STEM) images with high precision and robustness is a challenging task.