文献类型：期刊文献

英文题名：A Four-Task Convolutional Neural Network Model for Real-Time Volatile Organic Compounds Detection

作者：Cheng, Weiwei[1];Wang, Tao[2];Chen, Lechen[3];Ni, Wangze[3];Zhu, Jiaqing[1];Yang, Zhi[3];Xu, Shusheng[1];Zhang, Bowei[2];Xuan, Fuzhen[2]

机构：[1]Shanghai Univ Engn Sci, Sch Mat Sci & Engn, Shanghai 201620, 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]Shanghai Jiao Tong Univ, Sch Elect Informat & Elect Engn, Natl Key Lab Adv Micro & Nano Manufacture Technol, Shanghai 200240, Peoples R China

年份：2025

卷号：25

期号：15

起止页码：28568

外文期刊名：IEEE SENSORS JOURNAL

收录：;EI(收录号：20252718708973);WOS:【SCI-EXPANDED(收录号:WOS:001542426600002)】；

基金：This work was supported in part by the National Natural Science Foundation of China under Grant 62301314 and Grant 52321002 and in part by the In Situ Devices Center of East China Normal University.

语种：英文

外文关键词：Sensors; Training; Accuracy; Real-time systems; Data models; Predictive models; Intelligent sensors; Gas detectors; Convolutional neural networks; Feature extraction; electronic nose (E-nose); multitask deep learning; pattern recognition; real-time detection

摘要：The electronic nose (E-nose) is an advanced technique that has attracted substantial attention across various domains, such as environmental monitoring and disease prevention. This study proposes an innovative four-task convolutional neural network (FT-CNN) model to enhance the performance of E-nose. In contrast to conventional E-nose systems, the proposed FT-CNN model not only accomplishes four tasks simultaneously-gas classification, concentration prediction, state assessment, and anomaly identification-but also facilitates real-time detection, thereby enabling immediate decision-making in dynamic environments. The model employs a unique two-block knowledge-sharing structure, significantly improving system efficiency. Additionally, noise injection and data segmentation mechanisms are incorporated to bolster the robustness and generalization of the model. The million-volume dataset, collected from a self-developed gas sensing system, underscores the model's capacity to handle real-world variations and anomalies effectively. Experimental results demonstrate that the FT-CNN model achieves remarkable performance, with a gas classification accuracy of 97%, a sensor state assessment accuracy of 98%, an R-2 score exceeding 0.97 for concentration prediction, and an area under the curve (AUC) of 0.99 for anomaly identification. This comprehensive framework, integrating efficient data processing, noise immunity mechanisms, and real-time detection capabilities, demonstrates the remarkable potential of FT-CNN in advancing E-nose technology.