Mathematical model for predicting coffee roast degree using spectral reflectance characteristics from the AS7265X sensor

Some common methods for determining the coffee roast degree include colour measurement, moisture content analysis, and Agtron value assessment using a specialised spectrometer. This study develops a mathematical model to predict the roasting degree using a single reflectance characteristic from the AS7265X sensor. Spectral data from coffee samples were collected across different roasting stages, and mathematical models (i.e., linear, polynomial, exponential, logarithmic) with cross-validation techniques were built to establish relationships between individual reflectance features and key parameters: Agtron value, moisture content, and L* value. The results demonstrate that specific wavelengths strongly correlate with roasting degree indicators. The best performing models for each parameter were Agtron with reflectance at 860 nm using second order polynomial (R²cv= 0.985, RMSEcv = 4.671), Moisture content with reflectance at 460 nm using second order polynomial (R²cv = 0.838, RMSEcv = 0.864), and L* with reflectance at 810 nm using exponential (R²cv = 0.929, RMSEcv = 3.086). These findings highlight the potential of spectral data for non-destructive and real-time monitoring of the roasting process. The developed model offers an efficient tool for coffee quality control, particularly in industrial applications. Future research could explore machine learning techniques to enhance prediction accuracy using more robust and diverse datasets.