In Lithium-ion batteries, the graphite anode mainly contributes to the dilatation of the battery. Under the light microscope a distinct chromatic change can be observed during lithiation and de-lithiation due to the formation of graphite intercalation compounds (graphite: grey; LiC18: blue; LiC12: red; LiC6: gold). Using a novel cross-sectional in-situ optical microscopy setup for imaging of a Li-ion full cell with graphite anode and NMC 622 cathode, both effects can be studied simultaneously during charging and discharging. In this work we describe feature extraction methods to quantify these effects in the image data captured during the lithiation and de-lithiation process.
For the quantification of colorfulness, we evaluate different methods based on classical image processing. The metrics calculated with these approaches are compared to the results of ColorNet, which is a trainable colorfulness estimator based on deep convolutional neural networks. For the layer thickness measurement and the layer dilation derived from it, we propose a supervised semantic segmentation approach using U-Net.
All extracted features are evaluated based on their utility to develop data-driven models to predict different properties (like state of charge) directly from the image data during in-situ experiments.