Effect of coffee cherry maturity on the performance of the drying process of the bean: Sorption isotherms and dielectric spectroscopy

The study of the dynamic sorption isotherms of parchment coffee revealed that the maturity stage has an effect on the water availability of the bean, and that such dependence might affect the bean drying. The cultivar had an impact on the dispersion at high water activity values (aw > 0.8); hence, depicting dissimilar conditions before the drying is actually performed. The Guggenheim-Anderson-de Boer (GAB) sorption model revealed that stages 2, 3 and 7 present higher monolayer moisture content (Xₘ), K values close to one for all stages indicated a multilayer behaviour close to free water mobility and low C values for stages 2 and 7 represented a stronger water binding to the monolayer for these stages. Under static conditions and the drying conditions defined according to the moisture content of the beans, the unripe stages depicted an accelerated water activity decay that might promote stalling after roasting due to low values of the attribute (aw < 0.4). In a second phase of the study, the dielectric spectroscopy technique was used to evaluate the effect of the maturity and the drying category, and the response depended on the drying category rather than on the maturity stage except for the drying category with the higher moisture content range. The dielectric constant was utilized to predict both the moisture content and the water activity of the beans. The kernel-based ridge method with radial basis function for the moisture content and sigmoid function for the water activity model had good performance (R²: 0.82–0.88). A feature selection model based on a pipelined framework to include the whole spectra selected the 0.3 GHz to 0.9 GHz range as the optimal estimator, which was consistent with the performance of the single frequency models at 0.3 GHz. Although the errors were lower for the feature selection model, the single frequency is preferable as it has minimal complexity. As a non-linear dependence was evidenced, the machine learning strategies were necessary to respond to this condition. Consequently, the technology could be considered for the assessment of the water features of the bean during coffee drying or the storage of green coffee.