Respiratory system disease recognition based on diaphragm fiber-optic F-P sensor

• Published in: Sensors and actuators. A. Physical. Vol. 383; p. 116237
• Main Authors: Wang, Yangfeng, Wan, Shengpeng, Li, Yujie, Li, Zexin, Tang, Wenjun, Yu, Junsong
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2025
• Subjects: Deep Learning; Diaphragm fiber-optic F-P sensor; Optical fiber sensor; Respiratory system disease recognition; Diaphragm fiber-optic F-P sensor; Respiratory system disease recognition; Optical fiber sensor; Deep Learning
• ISSN: 0924-4247
• DOI: 10.1016/j.sna.2025.116237

In this paper, the respiratory system disease recognition technology based on diaphragm fiber-optic F-P(EFPI) sensor and CNN-BiLSTM network was studied, a beneficial attempt has been made to apply the membrane type fiber optic F-P cavity sensor in the recognition of respiratory diseases. The diaphragm fiber-optic F-P acoustic sensor was fabricated and used to collect respiratory signals for the test set. The training sets were derived from the ICBHI Challenge database, which consists of breath signals from 128 subjects. These subjects included healthy individuals and patients with respiratory diseases such as Asthma, Pneumonia, Bronchiectasis, COPD, URTI, etc. A CNN-BiLSTM model was established by combining convolutional neural network and bidirectional long and short term memory network, the training sets after using data augmentation methods are imported into this model for training, and the test sets were imported into the trained model for testing. Four types of sample sets, Bronchiectasis, COPD, URTI, and Health, were selected for recognition experiments, and the results show that the accuracy of F-P acoustic sensor based on CNN-BiLSTM model is 92.1 % in respiratory disease recognition test. Compared with traditional CNN model and LSTM model, the test accuracy of CNN-BiLSTM model is improved by 9.7 % and 11.9 %, respectively. In real life, we collecting 50 COPD respiratory records and 50 healthy respiratory records by the EFPI sensor, the prediction accuracy in the CNN-BiLSTM model is 96 % and 94 %, respectively. [Display omitted] •The membrane type fiber optic F-P cavity sensor is used to measure the respiratory signal.•A CNN-BiLSTM model is established to recognition of respiratory diseases.•Accurate identification of four respiratory diseases, including Bronchiectasis, COPD, URTI, and Health, has been achieved.