Olfaction, an essential human sense, has drawn significant attention for its sensitivity and environmental awareness, driving advancements in digital olfaction technologies. So far, the most widely used device in digital olfaction is the electronic nose, but direct contact with samples limits it applicability for widespread usage. In this paper, we bridge the gap by proposing Digital Olfactory Mucosa (DOM), which utilizes terahertz signals for contactless sensing and classifying odorant molecules by exploiting the molecular absorption properties of terahertz. Unlike recent wireless gas detecting works that focus on major molecules in the air, DOM specifically focuses on the rare odorant molecules in the air, usually less than 1ppm. To meet the challenge, we custom-design a unique signal processing model based on a hybrid neural network that can perform odorant feature sampling and fusion over time-domain and frequency-domain signals, so as to extract rare odorant molecular features. Furthermore, in order to better accommodate the diversity of odors, we design a dual granularity contrastive learning schema to optimize the spatial distribution of extracted olfactory features, which enhances the scalability of DOM. Evaluation on 25 odorant molecules within 10 odor clusters demonstrates DOM's ability to detect and classify odors, with 98.71% accuracy in odor cluster classification, and 98.66% in specific odor classification.
Lyu et al. (Sat,) studied this question.