Multi-task learning on microscopic hyperspectral data enables accurate classification of graphene oxide films.

The precise, high-throughput characterization of graphene oxide (GO) is essential for its use in advanced technologies but is hindered by its inherent structural heterogeneity. Traditional techniques, such as atomic force microscopy, are too slow for industrial quality control, while spectroscopic methods like Raman spectroscopy are often complicated by GO's structural disorder. To address these challenges, we propose a novel framework that integrates microscopic hyperspectral imaging (mHSI) with a multi-task learning (MTL) deep neural network architecture. The model was trained to classify GO films into three primary categories (single-layer, few-layer, and multi-layer) and twelve thickness-based subcategories, achieving exceptional classification accuracy of 97.1 % (12-class task) on an independent test set. The MTL deep learning model significantly outperformed a comparable single-task learning (STL) model, reducing the error rate by a factor of more than three. A control experiment using pseudo-RGB images underscored the critical role of hyperspectral data, with classification accuracy for the 12-class task dropping to 69.4 %, confirming mHSI's ability to capture essential spectral signatures. This work establishes a robust, data-driven paradigm for materials characterization, offering a scalable, non-destructive solution for advancing fundamental research and enabling high-throughput quality control in the production of GO-based devices.