Traceability of Green Tea Origin: An Adaptive Gas Features Classification Network Coupled With an Electronic Nose

Abstract

Green tea from different origins develops unique qualities and flavors due to varying environmental factors, such as climate, soil, and water quality. Unfortunately, lower quality green tea is sometimes misrepresented as coming from prestigious origins. This study presents a fast, objective, and effective gas detection method combined with deep learning to assess green tea quality from different origins. First, gas information from green tea of six renowned Chinese origins is captured using an electronic nose (e-nose) system. Next, we introduce an adaptive gas features calculation module (AGFCM) that integrates deep gas features through two methods: multiscales convolution calculations and adaptive attention mechanisms. Finally, we propose an adaptive gas features classification network (AGFC-Net) to classify the gas information from different origins. Following structural optimizations, ablation studies, and comparison across classification methods, AGFC-Net achieves the best results, with 98.42% accuracy, 98.56% ${F}_{{1}}$ -score, and 98.62% kappa coefficient. Overall, this e-nose-based gas detection technology, combined with AGFC-Net, enables effective and rapid identification of green tea quality variations, offering technical support for quality assurance and market safety.