基于D-GoogLeNet深度学习的呼气丙酮检测方法
Study of breath acetone detection based on D-GoogLeNet deep learning algorithm
李磊 1赵彦喆 1米玉泽 1朱宏殷1
作者信息
- 1. 长春工业大学电气与电子工程学院,吉林长春 130012
- 折叠
摘要
针对呼气丙酮检测提出了基于GoogLeNet深度学习框架结合电子鼻(E-nose)传感器阵列的非侵入式呼气检测方法.该方法不仅克服了传统电子鼻呼气检测在数据处理过程中需要手动提取特征的不足,还创新性将气体传感器时间序列响应数据通过可视化方法转换为响应图像,从而实现了混合气体中目标气体的准确识别;同时对现有的GoogLeNet架构进行了修改,改进后的模型(D-GoogLeNet)减少了过拟合现象的出现,即使在样本量较小的情况下也能实现有效分类;此外,为了验证模型的鲁棒性,在实验室模拟的患者不同浓度的呼气标志物中人为地引入高斯噪声,检验了模型的抗干扰能力.实验结果表明,在未添加噪声的情况下,丙酮和乙醇及其混合物的分类准确率、召回率和精确度均为1,当噪声标准差为100时,该模型对单一气体的分类准确率、精确度和召回率不受影响,仍然为1,但对混合物的分类准确度降为0.84,精确度和召回率降为0.94.实验结果证明了该检测方法的可行性,有望为临床检测奠定基础.
Abstract
For the detection of breath acetone in patients,this paper proposes a non-invasive breath detection method based on GoogLeNet deep learning framework combined with electronic nose(E-nose)sensors array.The proposed method not only overcome the traditional electronic nose breath testing shortcoming,manually extract the features in the process of data processing,but also innovatively converted the gas sensor response time series data into a response figure through image visualization methods,to realize the accurate identification of target gas in the gas mixture;at the same time,the existing GoogLeNet architecture was modified,the improved model(D-GoogLeNet)reduced the occurrence of overfitting phenomenon,and could achieve effective classification even with small sample sizes.In addition,in order to verify the robustness of the model,different concentrations of patients'breath markers are simulated,and the Gaussian noise is introduced,testing the anti interference ability of the model.The experimental results show that the classification accuracy,recall and precision of acetone,ethanol and their mixtures are all 1 without adding noise.When the standard deviation of noise is 100,the classification accuracy,precision and recall of the model for single gas is still 1,but the classification accuracy of the mixture is reduced to 0.84 and the precision and recall were reduced to 0.94.The experimental results proved the feasibility of the proposed method,which was expected to be a basis for clinical detection.
关键词
电子鼻/呼气检测/GoogLeNet/深度学习/噪声处理Key words
Electronic nose(E-nose)/breath detection/GoogLeNet/deep learning/noise preprocessing引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据