Multimodal spectral-imaging fusion with machine learning for non-destructive monitoring of Gardeniae fructus extract drying

Industrial drying of botanical extracts often suffers from variability due to empirical control. Here, we propose a real-time monitoring framework using Gardeniae fructus extract as a model system. A multimodal sensing system integrating RGB imaging and hyperspectral reflectance (350 – 2500 nm) was coupled with machine learning algorithms to monitor vacuum drying dynamics. Reference values for moisture content and chlorogenic acid (a key bioactive compound) were determined via standard methods. Spectral analysis identified the 1400 – 1900 nm region as critical for composition changes. Notably, fusing image texture features with spectral data significantly outperformed single-sensor approaches, achieving coefficients of determination of 0.99 for moisture and 0.96 for chlorogenic acid. This multimodal framework provides a rapid, non-destructive tool for process standardisation and quality control in functional food and nutraceutical production.