Visible Light Modulation of 16-Channel Sensor Arrays for Accurately Discriminating Volatile Sulfur Compounds at Room Temperature

Monitoring toxic volatile sulfur compounds (VSCs) by metal-oxide–semiconductor (MOS) sensors has gained much attention for various applications including smart factory and health screening. Although room temperature (RT) operation is highly preferred due to tiny power consumption and minimal explosion risk of VSCs, poor selectivity and insufficient molecule features acquired at RT pose a big challenge. Herein, a 16-channel MOS-based electronic nose (e-nose) has been integrated in a glass wafer with ITO interdigital electrode arrays, visible light modulation, and three kinds of feature extraction methods have been proposed to extract the (subtle) features of those VSCs molecules. Combining the transient e-nose response characteristics generated by visible light modulation and convolutional neural network (CNN) algorithm, a high prediction accuracy of 99.2% toward five kinds of VSCs with varying concentrations could be achieved. Furthermore, SHapley Additive exPlanations (SHAP) approach has been used to estimate the contribution of individual sensors in prediction models for the optimization of model computational complexity and sensor array size. This work sheds light on the rational screening of useful sensors for constructing high-performance e-nose with minimal costs for various applications.