Integrating convolutional neural networks with UV–Vis spectroscopy for monitoring the extraction procedure of Ganoderma lucidum spore powder

The rise of Ganoderma lucidum spores as a nutraceutical is gaining sharp acceptance and popularity. Aqueous extraction is the primary means of obtaining phytochemicals from the spores. However, low levels of phytochemicals and the lack of rapid, non-destructive detection methods present significant challenges for monitoring the extraction process. In this study, we combined ultraviolet–visible (UV–Vis) spectroscopy with convolutional neural network (CNN) to monitor the extraction process of sporoderm-broken Ganoderma lucidum spores (BGLS). Initially, the UV–Vis spectra of these samples were measured, and the levels of nine key ganoderic acids were determined using Ultra-Performance Liquid Chromatography-Triple Quadrupole Mass Spectrometry (UPLC-QqQ-MS). Subsequently, the spectra and analytes concentrations were employed as the input and output, for the CNN model, respectively. Finally, self-designed CNN model accurately characterized the changes in phytochemicals, with R 2 values great than 0.88 for all analytes. For all analytes, the CNN-based models without preprocessing were superior to other machine learning models in terms of prediction accuracy. This work demonstrates that combining UV–Vis with CNN is a feasible method for monitoring the extraction process of BGLS and advancing our understanding of the extraction process of the Traditional Chinese Medicines. • Combining CNN with UV–Vis spectra for real-time monitoring. • Self-designed CNN model outperformed classical machine learning models. • The end-to-end CNN model provided rapid guidance for industrial settings. • Uncovering the “black box” of the CNN model.