文献类型：期刊文献

英文题名：A Respiratory Disease Diagnosis Method of Electronic Nose Based on Pyramid Pooling and Self-Attention Mechanism

作者：Peng, Jingyi[1]; Mei, Haixia[1]; Wang, Tao[2]; Zeng, Min[3]; Xu, Dongdong[1]; Xing, Xiaoxue[1]; Shi, Lijuan[1]; Meng, Keyu[1]; Qin, Hongwu[1]; Zhao, Jian[1]; Zhang, Bowei[2]; Xuan, Fuzhen[2]

机构：[1] Changchun University, Key Lab Intelligent Rehabil & Barrier free Disable, Ministry of Education, Changchun, 130022, China; [2] School of Mechanical and Power Engineering, East China University of Science and Technology, Shanghai Key Laboratory of Intelligent Sensing and Detection Technology, Shanghai, 200237, China; [3] Shanghai Jiao Tong University, National Key Laboratory of Advanced Micro and Nano Manufacture Technology, Shanghai, 200240, China

年份：2024

外文期刊名：Proceedings of IEEE Sensors

收录：EI(收录号：20250317719673)

语种：英文

外文关键词：Contrastive Learning - Diagnosis - Lung cancer - Pulmonary diseases

摘要：In order to address the challenges of insufficient sensitivity and specificity, sample diversity, and unbalanced distribution of electronic noses(E-nose), this study proposes an innovative approach that integrates pyramid pooling and one encoder utilizing self-attention mechanisms for analyzing response data of E-nose. Pyramid pooling aggregates features across scales to capture multi-level global information, while a single encoder with self-attention mechanisms captures feature relationships. To enhance interpretability in deep learning models, SHapley Additive exPlanations analysis tools are used to clarify model explanations, aiding in understanding prediction logic and feature contributions. Through 3-fold cross-validation, the model achieved 92.31% accuracy, 94.84% precision, and 93.95% recall. This study provides an insightful approach to the implementation of E-nose systems for respiratory disease diagnosis. ? 2024 IEEE.