文献类型：期刊文献

英文题名：An Interpretable Sensor Selection Strategy for Array Optimization and Performance Enhancement

作者：Mei, Haixia[1];Peng, Jingyi[1];Wang, Tao[2];Zhang, Bowei[2];Xuan, Fuzhen[2]

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

年份：2025

卷号：10

期号：9

起止页码：6700

外文期刊名：ACS SENSORS

收录：;EI(收录号：20253919248774);WOS:【SCI-EXPANDED(收录号:WOS:001557018200001)】；

基金：This work was supported by the Jilin Provincial Natural Science Foundation (YDZJ202501ZYTS591), National Natural Science Foundation of China (62301314 and 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：E-nose; array optimization; SHAP; mutual information; interpretability

摘要：As more sensors are integrated into electronic nose (E-nose) systems to enhance gas detection accuracy, multitasking capabilities, and expand application scenarios, several challenges arise. These challenges include issues such as cross-sensitivity, increased hardware cost, heightened computational complexity, and information redundancy, due to the growing number of sensors in the array. As a result, array optimization plays a crucial role in improving the performance of multisensor systems. In this study, we propose a SHapley & Mutual Information-based Selection (SHMI-Select) method, which provides an interpretable sensor selection strategy for multisensor systems. This strategy initially selects the primary sensor based on interpretability analysis, followed by the identification of a secondary sensor through mutual information among sensors. Additionally, an incremental selection method is introduced to dynamically choose the optimal sensor combination, thereby ensuring system stability and adaptability in various tasks. Through interpretability analysis, our method not only helps to identify key sensors but also optimizes the sensor array combination. Finally, we validate the proposed array optimization method on three distinct E-nose data sets involving human breath, wine quality, and environmental gas. Compared to seven existing algorithms, our method substantially reduces sensor redundancy while boosting performance-achieving 62.5% fewer sensors with 10% accuracy gain on breath data, 83.3% reduction with 18% improvement on wine classification, and 62.5% reduction with a 2% R2 increase on environmental gas. The proposed method has good application value and economic benefits in the industrialization of future E-nose systems.