Machine learning methods for Schlieren imaging of a plasma channel in tenuous atomic vapor

TL;DR

This paper explores using machine learning, specifically deep neural networks trained on simulated data, to accurately analyze Schlieren images of plasma channels in atomic vapor, enabling precise measurement of plasma properties.

Contribution

It introduces a novel approach combining Schlieren imaging with deep learning to quantitatively analyze plasma channels in atomic vapor.

Findings

Neural networks can reliably extract plasma parameters from Schlieren images.

Deep learning models are resilient to slight experimental variations.

Simulated training data enables accurate parameter estimation in experiments.

Abstract

We investigate the usage of a Schlieren imaging setup to measure the geometrical dimensions of a plasma channel in atomic vapor. Near resonant probe light is used to image the plasma channel in a tenuous vapor and machine learning techniques are tested for extracting quantitative information from the images. By building a database of simulated signals with a range of plasma parameters for training Deep Neural Networks, we demonstrate that they can extract from the Schlieren images reliably and with high accuracy the location, the radius and the maximum ionization fraction of the plasma channel as well as the width of the transition region between the core of the plasma channel and the unionized vapor. We test several different neural network architectures with supervised learning and show that the parameter estimations supplied by the networks are resilient with respect to slight changes of the experimental parameters that may occur in the course of a measurement.