Authors: Patrick Austin (RAL, STFC), Keith Butler (SciML, STFC), Jeyan Thiyagalingam, and Tony Hey. *(Corresponding)
Electron microscopy has been undergoing revolution, with the advent of direct counting electron detectors providing very high detective quantum efficiency and low readout noise, facilitating very rapid frame collection rates.
Increasingly the importance of machine learning techniques is being recognised and exploited in materials’ science, for example identifying microstructure. Increased frame collection rates on transmission electron microscopes (TEM) allows the observation of dynamic processes such as defect migration and surface reconstruction. In these scenarios images are collected at a high frequency, machine learning techniques for rapid identification of features and objects within the image, such as semantic segmentation are already powerful tools.
Rapid machine learning, facilitated analysis and processing of images offers the promise of microscopes which automatically optimise data acquisition, or act as a monitor in nano-fabrication, or for alerting microscope operators of potentially important events.
In almost all instances it is desirable to have techniques to improve signal to noise ratios. For example, being able to image at lower electron doses can facilitate experiments with reduced beam induced phenomena taking place in samples; however these images are inevitably noisier than at higher doses. Denoising can facilitate low-dose experiments, with image quality comparable to high- dose experiments. Greater time resolution can be achieved with the aid of effective image denoising procedures.
We applied different deep-learning architectures in denoising TEM images of defective graphene sheets, assessing the performance in terms of three key considerations when deploying a denoising workflow:
- fidelity of the denoised image;
- speed of the algorithm;
- training data requirements
We are able to achieve denoising performance far in advance of the best available classical denoising techniques, opening the door for the application of deep-learning denoisng to facilitate enhanced electron microscopy.
virtualenv -p python3 ./env && source ./env/bin/activate && pip install mlcube-docker
git clone https://github.com/mlperf/mlcube_examples.git && cd ./mlcube_examples/emdenoise
# Configure EMDenoise MLCube
mlcube_docker configure --mlcube=. --platform=platforms/docker.yaml
# Run EMDenoise tasks: download data, preprocess data train and test the model
mlcube_docker run --mlcube=. --platform=platforms/docker.yaml --task=run/download.yaml
mlcube_docker run --mlcube=. --platform=platforms/docker.yaml --task=run/preprocess.yaml
mlcube_docker run --mlcube=. --platform=platforms/docker.yaml --task=run/train.yaml
mlcube_docker run --mlcube=. --platform=platforms/docker.yaml --task=run/test.yaml
Go to workspace/ directory and study its content.
Contributed by Digital Science Center, Indiana University Bloomington.