西安交通大学学报（医学版）2026，Vol.47Issue(2)：269-275,7.DOI:10.7652/jdyxb202602010
电子鼻漂移的图神经网络小样本补偿模型
A few-shot compensation model for electronic nose drift using graph neural networks
田垚 1张成 2王海容 3成诚4
作者信息
- 1. 西安交通大学未来技术学院,陕西 西安 710049
- 2. 西安交通大学机械工程学院,陕西 西安 710049
- 3. 西安交通大学机械工程学院,陕西 西安 710049||西安交通大学生物证据研究院/国家生物安全证据基地,陕西 西安 710049
- 4. 西安交通大学生物证据研究院/国家生物安全证据基地,陕西 西安 710049||西安交通大学国家卫生健康委法医学重点实验室,陕西 西安 710049
- 折叠
摘要
Abstract
Objective To address the sensor drift issue in electronic noses caused by temperature/humidity fluctuations in medical and outdoor scenarios,which leads to detection failures,this study proposes a few-shot compensation model.It resolves the bottleneck of traditional methods that rely on extensive drift data and struggle with long-term nonlinear drift adaptation.Methods We constructed the GNNSD model,integrating deep residual convolution and graph neural networks,as well as employing data augmentation and relational reasoning mechanisms,and conducted few-shot classification experiments on a public sensor drift dataset.Results The GNNSD model achieved an average accuracy of 84.12%under the K=1 setting,representing a 9.93%improvement over the best comparative algorithm,FEDA.Ablation experiments demonstrated the rationality of the model architecture.Conclusion By synergizing multi-scale feature extraction and graph-structured relational reasoning,the model maintains high classification accuracy even with only one reference sample per category.This provides a low-sample-dependent drift compensation solution for biosafety applications such as medical monitoring and cross-border screening.关键词
生物安全/电子鼻/传感器漂移/图神经网络(GNN)/小样本学习Key words
biological safety/electronic nose/drift of sensor/graph neural network(GNN)/small sample learning分类
机械制造引用本文复制引用
田垚,张成,王海容,成诚..电子鼻漂移的图神经网络小样本补偿模型[J].西安交通大学学报（医学版）,2026,47(2):269-275,7.
