Page 221 - 《应用声学》2023年第3期

第 42 卷 第 3 期 Vol. 42, No. 3
2023 年 5 月 Journal of Applied Acoustics May, 2023

⋄ 研究报告 ⋄

呼吸声分类技术研究及检测系统设计 ∗

张书文 侯 猛 刘泽华 张锦龙 †

(河南大学物理与电子学院 开封 475001)

摘要：患有呼吸疾病的病人由于呼吸道腺体不受意识控制、异物或者其本身肺部病变等原因会出现呼吸异常。
医护人员通过病人的呼吸状况可以找寻和分析病人出现呼吸异常的原因。以往呼吸声的诊断方式是专业的医
护人员通过使用听诊器对病人进行肺部听诊，但听诊的结果取决于医护人员的经验与相关参数。在初级诊断
治疗阶段，初级医生识别听诊声的准确率很低，一般从 20%∼80% 不等，因此会存在 10%∼20% 的高误诊率 (漏
诊、错诊和延误)。该文提出了一种利用聚偏二氟乙烯 (PVDF) 薄膜传感器提取语声特征的检测系统，该检测
系统根据病人发出声音的不同，提取呼吸声特征值判断病人的呼吸状况。通过 PVDF 传感器采集的微弱呼吸
声经过 K 最邻近分类之后的二分类平均识别率可达 96.56%，种类较多的四分类平均识别率为 89.84%。综上表
明，相比于传统听诊方式通过 PVDF 传感器采集识别的结果其具有更高的准确性和可靠性，其判断呼吸状况
的结果可以为医护人员提供参考，更好地监测病人的呼吸状况。
关键词：呼吸异常；薄膜传感器；声音特征；呼吸状态
中图法分类号: TN911.71 文献标识码: A 文章编号: 1000-310X(2023)03-0659-08
DOI: 10.11684/j.issn.1000-310X.2023.03.026

Research on respiratory sound classification technology and

detection system design

ZHANG Shuwen HOU Meng LIU Zehua ZHANG Jinlong

(School of Physics and Electronics, Henan University, Kaifeng 475001, China)

Abstract: Patients with respiratory diseases may experience breathing abnormalities due to unconsciousness
of respiratory glands, foreign matter or their own lung lesions. Medical staff can find and analyze the causes
of the patient’s abnormal breathing through the patient’s respiratory condition. In the past, the diagnosis
method of respiratory sound was that professional medical staff auscultated the patient’s lungs by using
stethoscope. However, the result of auscultation depends on the experience and relevant parameters. In the
primary diagnosis and treatment stage, the accuracy of primary doctors in identifying auscultation sounds is
very low, generally ranging from 20% to 80%. Therefore, there will be a high misdiagnosis rate of 10% to 20%
(missed diagnosis, wrong diagnosis and delay diagnosis). In this paper, a detection system using polyvinylidene
difluoride (PVDF) thin film sensor to extract speech features is proposed. According to the different sounds of
patients, the eigenvalues of sounds are extracted to judge the respiratory status of patients. After K-nearest
neighbor classification, the average recognition rate of the two classification of the weak respiratory sounds col-
lected by by PVDF sensor can reach 96.56%, and the average recognition rate of the four classification with more

2022-02-24 收稿; 2022-09-23 定稿
国家自然科学基金面上项目 (61675063, 61975052), 河南大学一流学科培育项目 (2019YLZD06), 河南省科技攻关项目 (162102210021)
∗
作者简介: 张书文 (1996– ), 男, 河南周口人, 硕士研究生, 研究方向: 生物声学及光电传感器。
† 通信作者 E-mail: zjl@henu.edu.cn