Convolutional neural network approach to ion Coulomb crystal image analysis

Abstract

This paper reports on the use of a convolutional neural network methodology to analyze fluorescence images of calcium-ion Coulomb crystals in the gas phase. A transfer-learning approach is adopted using the publicly available RESNET50 model. It is demonstrated that by retraining the neural network on around 500 000 simulated images, we are able to determine ion-numbers not only for a validation set of 100 000 simulated images but also for experimental calcium-ion images from two different laboratories using a wide range of ion-trap parameters. Absolute ion numbers in the crystal were determined for the experimental data with a percentage error of ∼10%. This analysis can be performed in a few seconds for an individual crystal image, and therefore, the method enables the objective, and efficient, analysis of such images in real time. The approach adopted also shows promising performance for identifying Ca+ ion numbers in images of mixed-species crystals, thereby enhancing the experimental methodologies for studying the kinetics and dynamics of cold ion–molecule reactions.