文献类型：期刊文献

中文题名：基于膨胀因果卷积和注意力机制的气体识别方法

英文题名：Gas Recognition Method Based on Dilatation Causal Convolution and Attention Mechanism

作者：俞凌伟[1];杨孟平[1];杨海[1];王喆[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2025

卷号：51

期号：3

起止页码：380

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

基金：国家重点研发计划项目(2022YFB3203500)。

语种：中文

中文关键词：电子鼻系统;气体识别;注意力机制;时态卷积网络;时间序列

外文关键词：electronic nose system;gas identification;attention mechanism;temporal convolutional networks;time series

摘要：提出了一种基于膨胀因果卷积和注意力机制的气体识别方法,该算法结合Transformer中的注意力机制和多尺度时态卷积网络提取全局和局部特征,获得了更具表示性的特征和更大的感受野,捕获气体的瞬时信息和变化趋势。在Open Sampling、Drift、Twin3个不同数据集上进行了实验,结果表明,所提出的方法分别达到99.47%、99.61%和99.22%的准确率,优于现有主流方法,证实了其有效性。
Gas identification is of great significance in the fields of environmental monitoring,industrial safety and medical health,which can effectively detect harmful gas leaks,monitor air quality and identify disease odor markers.However,the field of gas identification is faced with the problem that the sensor data needs to be processed manually before it can be used for subsequent analysis.Convolutional neural network(CNN)has been gradually applied in gas recognition scenarios of electronic nose systems with their ability of automatic feature learning and endto-end modeling.Although CNN performs well in this field,there are still challenges such as limited receptive field and insufficient global feature extraction,resulting in limited recognition performance.To solve these problems,a gas recognition method based on expansive causal convolution and attention mechanism is proposed.The algorithm combines the attention mechanism and multi-scale temporal convolution network in Transformer to extract global and local features,extract more representational features and obtain a larger receptive field,and capture the instantaneous information and change trend of gas.Experiments are conducted on three different data sets—Open Sampling,Drift and Twin.The results show that the proposed method achieves accuracy of 99.47%,99.61%and 99.22%,respectively,which are superior to the existing mainstream methods,thereby confirming its effectiveness.