Sweet-honey aroma intensity in tobacco leaves under different flue-curing temperature-humidity regimes: Rapid discrimination using an electronic nose-1D-CNN model

This study investigated the association between temperature–humidity regulation during flue-curing and sweet-honey aroma intensity and developed a rapid discrimination approach based on electronic-nose responses. Sensory evaluation showed significant differences among curing regimes, although substantial within-regime variation indicated additional effects of raw-material differences. Mainstream-smoke volatile profiles showed only weak grade-related differences, and associations of individual compounds with sensory grades were not significant after multiple-testing correction. Individual sensor responses and linear multichannel analyses also showed limited discriminatory ability. Using the same five-grade framework, RF, RBF-SVM, and a one-dimensional convolutional neural network (1D-CNN) were compared. The 1D-CNN achieved 92.0% test accuracy (macro-F1 = 0.888), compared with 80.0% for RF and 84.0% for RBF-SVM. Three of four independent external samples were correctly classified. These findings support the potential of electronic-nose fingerprinting combined with nonlinear classification for rapid evaluation of sweet-honey aroma intensity, pending validation in larger and more diverse datasets.