Dataset Information

A small-dataset-trained deep learning framework for identifying atoms on transmission electron microscopy images.

ABSTRACT: To accurately identify atoms on noisy transmission electron microscope images, a deep learning (DL) approach is employed to estimate the map of probabilities at each pixel for being an atom with element discernment. Thanks to a delicately-designed loss function and the ability to extract features, the proposed DL networks can be trained by a small dataset created from approximately 30 experimental images, each with a size of 256 × 256 pixels². The accuracy and robustness of the network were verified by resolving the structural defects of graphene and polar structures in PbTiO₃/SrTiO₃ multilayers from both the general TEM images and their imitated images on which intensities of some pixels lost randomly. Such a network has the potential to identify atoms from very few images of beam-sensitive material and explosive images recorded in a dynamical atomic process. The idea of using a small-dataset-trained DL framework to resolve a specific problem may prove instructive for practical DL applications in various fields.

SUBMITTER: Chen Y

PROVIDER: S-EPMC9929221 | biostudies-literature | 2023 Feb

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

A small-dataset-trained deep learning framework for identifying atoms on transmission electron microscopy images.

Chen Yuan Y Liu Shangpeng S Tong Peiran P Huang Ying Y Tian He H Lin Fang F

Scientific reports 20230214 1

To accurately identify atoms on noisy transmission electron microscope images, a deep learning (DL) approach is employed to estimate the map of probabilities at each pixel for being an atom with element discernment. Thanks to a delicately-designed loss function and the ability to extract features, the proposed DL networks can be trained by a small dataset created from approximately 30 experimental images, each with a size of 256 × 256 pixels<sup>2</sup>. The accuracy and robustness of the networ ...[more]

PMID: 36788257

Similar Datasets

Project description:For many reaction processes, such as catalysis, phase transformations, nanomaterial synthesis etc., nanoscale observations at high spatial (sub-nanometer) and temporal (millisecond) resolution are required to characterize and comprehend the underlying factors that favor one reaction over another. The combination of such spatial and temporal resolution (up to 600 µs), while rich in information, produces a large number of snapshots, each of which must be analyzed to obtain the structural (and thereby chemical) information. Here we present a methodology for automated quantitative measurement of real-time atomic position fluctuations in a nanoparticle. We leverage a combination of several image processing algorithms to precisely identify the positions of the atomic columns in each image. A geometric model is then used to measure the time-evolution of distances and angles between neighboring atomic columns to identify different phases and quantify local structural fluctuations. We apply this technique to determine the atomic-level fluctuations in the relative fractions of metal and metal-carbide phases in a cobalt catalyst nanoparticle during single-walled carbon nanotube (SWCNT) growth. These measurements provided a means to obtain the number of carbon atoms incorporated into and released from the catalyst particle, thereby helping resolve carbon reaction pathways during SWCNT growth. Further we demonstrate the use of this technique to measure the reaction kinetics of iron oxide reduction. Apart from reducing the data analysis time, the statistical approach allows us to measure atomic distances with sub-pixel resolution. We show that this method can be applied universally to measure atomic positions with a precision of 0.01 nm from any set of atomic-resolution video images. With the advent of high time-resolution direct detection cameras, we anticipate such methods will be essential in addressing the metrology problem of quantifying large datasets of time-resolved images in future.

Dataset Information

A small-dataset-trained deep learning framework for identifying atoms on transmission electron microscopy images.

Publications

A small-dataset-trained deep learning framework for identifying atoms on transmission electron microscopy images.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets