Development of an Edge-AI-based Electronic Nose for Quality Assessment of Arabica Coffee

Arabica coffee quality is conventionally evaluated using sensory cupping, a subjective, timeconsuming, and highly dependent method on expert panelists. To overcome these limitations, this study proposes an electronic nose system integrated with a one-dimensional CNN for objective and nondestructive classification of Arabica coffee quality based on aroma characteristics. To regulate volatile organic compound exposure, the system employs eight metal-oxide semiconductor gas sensors installed in a controlled sensing chamber, supported by diaphragm pumps and solenoid valves. Aroma acquisition was conducted over a 300-s sensing cycle consisting of baseline, exposure, and purging phases, producing 72,000 time-series data points from 240 experimental trials, with each trial representing a single sample. Different sensor response patterns were observed between the quality classes. High-grade Arabica coffee samples exhibited voltage responses ranging from 700 to 4500 mV, whereas low-grade samples showed responses between 500 and 4000 mV, indicating differences in volatile emission profiles. A 1D-CNN model was developed for classification using 300 samples, with 192 samples allocated for training and 48 for validation. The proposed model achieved 100% validation accuracy, with no misclassification observed in the confusion matrix and precision, recall, and F1-score values equal to 1.0. Training evaluation further demonstrated an overall accuracy of 99.47%, with precision, recall, and F1-score values of 0.962, 0.959, and 0.959, respectively. These findings demonstrate that the proposed E-Nose system, combined with a CNNbased classifier, provides a fast, robust, and objective framework for assessing the quality of Arabica coffee and shows strong potential as an alternative to conventional sensory evaluation.