Coffee is consumed by over one billion people daily, creating a high risk of adulteration where premium varieties are blended with inferior fillers. While various analytical techniques exist, they are often expensive and require labor-intensive sample preparation, such as grinding and brewing. Here, we present a proof-of-concept of a non-destructive platform that integrates ion-mobility spectrometry with online hot-gas extraction for rapid classification of coffee varieties using single-bean samples. The integration of a 1D convolutional neural network (CNN) model enables automated differentiation with 100% classification accuracy across four coffee bean varieties (Civet Arabica, Civet Robusta, carbonic maceration, and anaerobic fermentation natural) at an optimal bin width of 0.25 ms. When expanded to 10 diverse coffee bean varieties, the 1D CNN model achieved ∼92% independent test accuracy at a bin width of 0.10 ms. Furthermore, the system successfully monitored aroma degradation over 3 weeks and accurately predicted degradation patterns of anaerobic fermentation natural Arabica. Additionally, this 1D CNN model achieved 90% accuracy in predicting the presence of adulterants in coffee beans. Although the proposed method does not identify specific VOCs, the heuristic readouts it generates still enable rapid quality control and authenticity verification in commercial coffee beans.
Prayoga et al. (Thu,) studied this question.