Neural networks can be trained to remove noise that affects the measurement of quantitative properties of high-energy density X-ray images.
Each step in a high energy density (HED) experiment adds noise to the resulting X-ray image, which affects the estimation of properties of the imaged material, making noise removal as important as the imaging experiment itself. Based on a combination of X-ray source and detector effects, noise can be present in different forms and amounts, so it is important for researchers to find noise removal methods that are tailored to their data.
Levesque and his colleagues particularly needed a way to remove the noise that affects small-scale fluctuations. By training a neural network model, they were able to significantly reduce noise in experiments designed to study Rayleigh-Taylor and Richtmyer-Meshkov instabilities.
“We can train our model using a set of natural images that have been split into small patches and corrupted with a simple noise model,” says author Joseph Levesque. “The network architecture combines existing ideas from the field of image reconstruction with machine learning with modifications selected to better represent the noise contribution in the data.”
The denoiser is trained to remove a wide range of noise estimated from the team's data, and its performance on test images gives confidence in its applicability to the data. Once trained, the model can be easily applied to any image without additional tuning parameters. If the noise in your system falls outside the range of your model, it is relatively easy to train a new model with the same architecture and methodology.
“In principle, by combining this general model with a more accurate forward noise model, we could eventually eliminate noise in nearly all x-ray-based imaging,” Levesque said.
sauce: “Neural Network Denoising of X-Ray Images from High Energy Density Experiments,” by Joseph M. Levesque, Elizabeth C. Merritt, Kirk A. Flippo, Alexander M. Rasmus, and Forrest W. Doss, Scientific Equipment Reviews (2024). You can find this article here: https://doi.org/10.1063/5.0207005 .
