文献类型：期刊文献

英文题名：Smart VOCs recognition system based on single gas sensor and multi-task deep learning model

作者：Mei, Haixia[1];Peng, Jingyi[1];Wang, Tao[2];Zhang, Bowei[2];Xuan, Fuzhen[2];Wang, Lijie[3];Zeng, Min[4];Yang, Zhi[4]

机构：[1]Changchun Univ, Key Lab Intelligent Rehabil & Barrier free Disable, Minist Educ, Changchun 130022, Peoples R China;[2]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai Key Lab Intelligent Sensing & Detect Tech, Shanghai 200237, Peoples R China;[3]Jilin Univ, Coll Elect Sci & Engn, State Key Lab Integrated Optoelect, Changchun 130012, Peoples R China;[4]Shanghai Jiao Tong Univ, Sch Elect Informat & Elect Engn, Dept Micro Nano Elect, Natl Key Lab Adv Micro & Nano Manufacture Technol, Shanghai 200240, Peoples R China

年份：2025

卷号：439

外文期刊名：SENSORS AND ACTUATORS B-CHEMICAL

收录：;EI(收录号：20251718293507);WOS:【SCI-EXPANDED(收录号:WOS:001491911000001)】；

基金：This work was supported by the Jilin Provincial Science and Technology Development Program Project (YDZJ202501ZYTS591) , National Natural Science Foundation of China (62301314) , and the Science Fund for Creative Research Groups of the National Natural Science Foundation of China (52321002) .This work also received support from the In Situ Devices Center of East China Normal University.

语种：英文

外文关键词：Electronic nose; Temperature modulation; Transfer learning; Multi-task model; Anomaly detection; MEMS gas sensor

摘要：Electronic nose (E-nose) technology faces key challenges, including cross-sensitivity in gas sensors, high power consumption in multi-sensor arrays, and the need for scalable models. Additionally, robust detection and calibration of abnormal samples are essential for practical sensor applications. Looking forward, achieving costeffectiveness, long-term stability, and enhanced selectivity and sensitivity of metal oxide semiconductor gas sensors is crucial for E-nose commercialization. In this study, a temperature modulation technique is employed to capture the response from a single MEMS sensor to 10 volatile organic compounds, achieving an average gas identification accuracy of 98.38 % using a convolutional neural network. Transfer learning was also applied to determine gas concentrations, achieving an average accuracy of 88.65 %. To address the interference from the sensor itself or environmental factors, an autoencoder was trained to detect and compensate for anomalies, relying exclusively on features from normal samples and thereby minimizing the need for dedicated abnormal datasets. Anomaly detection accuracy reached 99.00 %. Overall, this research provides valuable insights toward developing low-power, multitasking, high-performance E-nose technology, advancing its potential for real-world applications.