Image‐Based Quantification of Deposited Lithium on Li Metal Surfaces

Image‐based quantification of the amount of deposited lithium on Li metal surfaces remains a big challenge. This work introduces an automated workflow consisting of the following steps: pixel‐wise identification of deposited lithium, segmentation of pixel clusters in instances of deposited Li structures, and instance‐based quantification of deposited lithium. Two algorithms for the identification of deposited lithium are developed. The rule‐based approach uses a combination of logical filters to detect deposited lithium. Core element of the neural network‐based approach is a single convolution neural network. Images of deposited lithium on Li electrodes in a Li||Li in situ cell are used as input data. Selected images are manually masked to generate data to optimize, train, and evaluate the performance of both image processing approaches. Segmentation enables to quantify instances of deposited Li structures by number, characteristic length scale, and volume. Automated image processing makes the assessment of the total deposited volume of lithium at the end of a constant current phase feasible, as well as the evolution of the total deposited volume during the constant current phase for various current densities.