Electroluminescence imaging, a widely utilized technique in solar cell analysis, is often employed for mapping optoelectronic parameters. Voltage-dependent Electroluminescence (ELV) measurements have been shown to mimic local diode current-voltage characteristics [1]. A corresponding physical model enables the derivation of two local parameters from ELV data measured on solar cells: a pseudo-recombination current J0∗ and a pseudo-series resistance Rs∗. Various local characteristics of the cells, namely the series resistance and dark saturation current, can be deduced from these parameters. ELV measurements performed on solar cells are stored in large data cubes, typically sized at a few hundred thousand pixels. Pixel-wise regression of this data is possible through Least Squares minimization; however, this method is time-consuming and requires a trade-off between data dimension, fitting accuracy, and computation duration. To mitigate the need for compromise, we propose using Machine Learning (ML) techniques, known for their efficiency in rapidly processing large datasets. Additionally, the knowledge of a physical model allows for generating a significant amount of ELV data numerically, enabling supervised training of ML models. We initially employ a Multilayer Perceptron (MLP) for pixel-wise analysis of the data cube, utilizing an ELV curve as input to predict either Rs∗ or J0∗. Secondly, we use a fully Convolutional Neural Network (CNN), known as U-NET [2], to process the entire cube as input, generating a parameter map. The MLP is almost as accurate as the Least Squares fitting, whereas the CNN prediction precision is lower. Compared to Least Squares fitting, the use of ML permits a significant reduction in analysis duration—by a factor of 240 (using the MLP) to 1200 (using the CNN) in this study; this paves the way for real-time analysis of solar cells. This technique could be applied to study other types of data cubes, such as those resulting from Hyperspectral imaging. [1] D. Ory, N. Paul, and L. Lombez, ‘Extended quantitative characterization of solar cell from calibrated voltage-dependent electroluminescence imaging’, Journal of Applied Physics, vol. 129, no. 4, p. 043106, Jan. 2021, doi: 10.1063/5.0021095. [2] O. Ronneberger, P. Fischer, and T. Brox, ‘U-Net: Convolutional Networks for Biomedical Image Segmentation’. arXiv, May 18, 2015. Accessed: Jan. 26, 2024. [Online]. Available: http://arxiv.org/abs/1505.04597