Kong Ruiwen, Shen Wenfeng, Gao Yang, Lv Dawu, Ai Ling, Song Weijie, Tan Ruiqin
Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo 315211, China.
Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing 100049, China.
Sensors (Basel). 2025 Feb 28;25(5):1480. doi: 10.3390/s25051480.
This article introduces a novel approach to improve electronic nose classification accuracy by optimizing sensor arrays and aligning features. This involves selecting the best sensor combinations and reducing redundant information for better odor recognition. We employ a feature alignment algorithm to address the discrepancies that impede model sharing between electronic nose devices. Our research focuses on overcoming challenges associated with material selection and the constraints of transferring classification models across different electronic nose devices for drug classification. We fabricated six SnO-based MEMS gas sensors using physical vapor deposition. The ReliefF algorithm was employed to rank and score each sensor's contribution to drug classification, identifying the optimal sensor array. We then applied feature alignment from transfer learning to enhance model sharing among three inconsistent devices. This study resolves the issue of electronic noses being hard to use on the same database due to hardware inconsistencies in batch production, laying the groundwork for future mass production.
本文介绍了一种通过优化传感器阵列和对齐特征来提高电子鼻分类准确率的新方法。这包括选择最佳的传感器组合并减少冗余信息以实现更好的气味识别。我们采用一种特征对齐算法来解决阻碍电子鼻设备之间模型共享的差异问题。我们的研究重点是克服与材料选择相关的挑战以及在不同电子鼻设备之间传输用于药物分类的分类模型时所面临的限制。我们使用物理气相沉积法制造了六个基于SnO的MEMS气体传感器。采用ReliefF算法对每个传感器对药物分类的贡献进行排名和评分，从而确定最佳的传感器阵列。然后，我们应用迁移学习中的特征对齐来增强三个不一致设备之间的模型共享。这项研究解决了由于批量生产中的硬件不一致导致电子鼻难以在同一数据库上使用的问题，为未来的大规模生产奠定了基础。
