摘要
提出一种基于卷积神经网络的便携式智能仿生嗅觉微机电系统(micro-electro-mechanical system,MEMS)设计方法。采用高集成度异质标记物敏感单元阵列作为系统前端气体感知器,通过RMSprop优化算法构建一维卷积神经网络丙酮气体质量浓度检测模型,可实现模拟多组分气体环境中丙酮气体的高精度定量分析。实验结果表明,该模型的拟合优度为0.995,平均相对误差为0.0885,平均绝对误差为0.1216 mg/m 3,检测效果优于多层感知器神经网络、支持向量回归机等其他6种算法模型。在基于该系统开展的呼气实验测试中,15组样本的检测标准差在0.0125~0.3709 mg/m 3之间,具有优异的稳定性。
A method to design the portable intelligent bionic olfactory micro-electro-mechanical system(MEMS)based on convolutional neural network was proposed.A highly integrated array of heterogeneous marker sensitive units was used as the gas sensing device.High-precision quantitative online analysis of acetone gas in a simulated multi-component gas environment could be realized through one-dimensional convolutional neural network acetone gas concentration detection model based on RMSprop optimization algorithm.The experimental results show that the detection effect of one-dimensional convolutional neural network model is better than that of other six models such as multi-layer perceptron neural network and support vector regression machine.The one-dimensional convolutional neural network model has a goodness of fit of 0.995,an average relative error of 0.0885,and an average absolute error of 0.1216 mg/m 3.In the exhalation test based on the system,the detected standard deviation of the 15 sample groups ranges from 0.0125 to 0.3709 mg/m 3,indicating excellent stability.
作者
周雨萌
尤睿
魏向阳
祝连庆
ZHOU Yumeng;YOU Rui;WEI Xiangyang;ZHU Lianqing(MOE Key Laboratory of Optoelectronic Measurement Technology and Instrument,Beijing Information Science&Technology University,Beijing 100192,China;Beijing Laboratory of Biomedical Testing Technology and Instruments,Beijing 100192,China)
出处
《北京信息科技大学学报(自然科学版)》
2023年第3期52-58,共7页
Journal of Beijing Information Science and Technology University
基金
国家重点研发计划(2021YFB3201303)
山东省重点研发计划(2022CXPT045)
生物医学检测技术及仪器北京实验室项目(100303002)。
关键词
仿生嗅觉
丙酮气体检测
微机电系统
卷积神经网络
biomimetic olfactory
acetone gas detection
micro-electro-mechanical system(MEMS)
convolutional neural network