[1]林冬梅,王鑫瑜,杨玉荷,等.基于EfficientNetV2-多注意力机制的脉象识别[J].计算机技术与发展,2025,(10):173-180.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0134]
　LIN Dong-mei,WANG Xin-yu,YANG Yu-he,et al.Pulse Recognition Based on EfficientNetV2-multiattention Mechanism[J].,2025,(10):173-180.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0134]
点击复制

基于EfficientNetV2-多注意力机制的脉象识别

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年10期
页码:
173-180
栏目:
新型计算应用系统
出版日期:
2025-10-10

文章信息/Info

Title:
Pulse Recognition Based on EfficientNetV2-multiattention Mechanism
文章编号:
1673-629X(2025)10-0173-08
作者:
林冬梅1; 王鑫瑜1; 杨玉荷1; 陈扶明2; 丑永新3
1. 兰州理工大学 电气工程与信息工程学院,甘肃兰州 730050; 2. 中国人民解放军联勤保障部队第九四〇医院医疗保障中心,甘肃兰州 730030; 3. 常熟理工学院 电气与自动化工程学院,江苏常熟 215506
Author(s):
LIN Dong-mei1; WANG Xin-yu1; YANG Yu-he1; CHEN Fu-ming2; CHOU Yong-xin3
1. School of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China; 2. Medical Support Center,the 940th Hospital of Joint Logistics Support Force of Chinese People's Liberation Army,Lanzhou 730030,China; 3. School of Electrical and Automation Engineering,Changshu Institute of Technology,Changshu 215506,China
关键词:
脉象信号识别; 格拉姆角场(GAF); 二维图像; EfficientNetV2; 多注意力机制
Keywords:
pulse signal recognition; Gram angle field (GAF); 2D image; EfficientNetV2; multiattention mechanism
分类号:
TP391.4
DOI:
10.20165/j.cnki.ISSN1673-629X.2025.0134
摘要:
脉象信号识别分析中,时域、频域以及时频域等方法难以将脉象信号的非线性特征进行深度提取,并且会造成运算缺失和数据丢失,无法进行特征的自学习。针对以上问题,该文提出了一种基于EfficientNetV2-多注意力机制的脉象信号识别分析方法。相较以往使用特征进行脉象识别的方法,二维图像的脉象识别完整保留了脉象信号的特征信息,提高了识别的准确率。该文利用桡动脉模拟平台和双目视觉脉搏采集系统,模拟和采集脉象信号。将采集到的薄膜图像经过一系列处理后得到5种脉象的多维脉搏波,将16维20周期的数据进行周期划分,扩增一维数据集,使用格拉姆角场(GAF)将一维脉象序列转换为二维图像,最后通过EfficientNetV2-多注意力机制网络模型进行脉象信号的识别。实验结果表明,该方法进行脉象识别的平均准确率可达到98.4%,其中促脉和浮脉的识别准确率可达100%,且相较于其他卷积神经网络,该模型有较高的准确率、较短的训练时间和较小的网络参数量。
Abstract:
In pulse signal recognition analysis,time domain,frequency domain and time-frequency domain methods are difficult to extract the nonlinear features of pulse signals in depth,and cause missing operations and data loss,and are unable to perform self-learning of fea-tures. To address these problems,an EfficientNetV2-based multi-attention mechanism for pulse signal recognition and analysis is proposed. Compared with the previous methods of pulse recognition using features,the pulse recognition of 2D images completely retains the feature information of pulse signals and improves the recognition accuracy. In this paper,a radial artery simulation platform and a bin-ocular vision pulse acquisition system are used to simulate and acquire pulse signals. The acquired thin-film images are processed in a series of ways to obtain multidimensional pulse waves of five pulse types. Divide the 16-dimensional 20-cycle data into cycles,augment the one-dimensional dataset,convert the one-dimensional pulse sequences into two-dimensional images using Gram's angle field (GAF),and then recognize pulse signals using the EfficientNetV2-multi-attention mechanism network model. The experimental results show that the average accuracy for pulse recognition can reach 98.4%,among which the recognition results of promoting pulse and floating pulse can reach 100%. Compared with other convolutional neural networks,this model has higher accuracy,shorter training time and smaller number of network parameters.
更新日期/Last Update: 2025-10-10