基于卷积神经网络模块化搜索的高效电子鼻多气体分类算法

Efficient Electronic Nose Based on Convolutional Neural Network Modular Search Multi Gas Classification Algorithm

祝煜荻 ¹曾敏 ¹杨建华 ¹胡南滔 ¹杨志¹

扫码查看

作者信息

1. 上海交通大学,上海 200240
折叠

摘要

该文设计了一种基于格拉姆角和场的传感器信号转图方法,并提出了一种基于AlexNet的卷积神经网络模块化结构搜索方法(block-GS).实验结果表明,block-GS方法能够搜索到性能优秀的网络结构,在两个气体数据集上的分类准确率分别达到92.11%和93.33%,比普通网格搜索提高了近5%.此方法有望成为电子鼻模式识别算法设计的有效解决途径之一.

Abstract

This paper designs a sensor signal mapping method based on Gram angle and field,and proposes a convolutional neural network modular structure search method(block GS)based on AlexNet.The experimental results show that the block GS method can search for high-performance network structures,with classification accuracies of 92.11%and 93.33%on two gas datasets,respectively,which is nearly 5%higher than ordinary grid search.This method is expected to become one of the effective solutions for the design of electronic nose pattern recognition algorithms.

关键词

电子鼻/格拉姆角和场/卷积神经网络/网格搜索/气体分类算法

Key words

electronic nose/graham point and field/convolutional neural network/grid search/gas classification algorithm

引用本文复制引用

出版年

2024

数字通信世界

电子工业出版社

数字通信世界

影响因子：0.162

ISSN：1672-7274

段落导航