基于温度调制与边缘计算的智能电子鼻系统研究  

Study of Intelligent Electronic Nose System Based on Temperature Modulation and Edge Computing

作　　者：王涛 吴羽 张恺悦 金玉山 倪望泽 张永威 杨志[1] WANG Tao;WU Yu;ZHANG Kaiyue;JIN Yushan;NI Wangze;ZHANG Yongwei;YANG Zhi(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Marine Diesel Engine Research Institute,Shanghai 201108,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]中国船舶集团有限公司第七一一研究所,上海201108

出　　处：《无人系统技术》2023年第5期28-39,共12页Unmanned Systems Technology

基　　金：国家自然科学基金(62301314);中国博士后科学基金(2023M732198)。

摘　　要：针对传统电子鼻系统中通常需要较多数量的传感器构成传感阵列,导致体积、功耗、成本急剧上升的问题,提出了传感阵列虚拟化的方法。首先通过周期性地调整传感器工作温度获得高频响应信号,利用包络线特征提取算法将高频信号分解为可用于模式识别的多个低频响应信号;其次通过设计适用于ARM Cortex-M3内核单片机上运行的人工神经网络模型,在气体传感模块上部署了模式识别系统,增强了系统的高内聚/低耦合特性;最终经验证,构建的智能气体传感系统可以实现对8种可挥发有机蒸气类别与浓度的预测,分类的平均准确率为91%,浓度回归的平均R2分数为0.83。与传统常温加热模式的实体传感器阵列相比,通过脉冲加热调制构建的智能气体传感系统将功耗降低38%,分类准确率提高了12%,显著提升了电子鼻系统的识别性能和经济实用性。In traditional electronic nose systems,a large number of sensors are used to construct a sensor array,which leads to dramatic increase of volumes,power consumption,and cost.To solve this problem,this paper has proposed a sensor array virtualization method.Firstly,a high-frequency response signal is obtained by periodically adjusting the working temperature of the sensor,which is decomposed into multiple low-frequency response signals by using the envelope feature extraction algorithm.Then,the pattern recognition system is deployed on the gas sensing module by designing an efficient artificial neural network model.This model is suitable for running on the ARM Cortex-M3 based microcontroller and enhances the high cohesion/low coupling characteristics of the system.Finally,the intelligent gas sensing system constructed in this work can realize the category classification and concentration prediction of 8 kinds of Volatile Organic Compounds after verification.The average accuracy of classification is 91%,and the average R2 score of regression is 0.83.Compared with the traditional sensor array,the power consumption can be reduced by 38%and the classification accuracy can be increased by 12%with the newly designed system,which significantly improves the recognition performance and economic performance of the electronic nose system.

关 键 词：气体传感器 传感阵列 电子鼻 温度调制 边缘计算 人工神经网络 

分 类 号：TH832[机械工程—仪器科学与技术]