Intelligent Respiratory Status Monitoring via a 1DCNN-Assisted Microfiber Sensor

Respiration is a crucial parameter for monitoring human health. Traditional respiration sensors are plagued by complex manufacturing procedures, unpleasant wearing, limited accuracy, and the requirement for meticulous alignment. To address these issues, we suggest using a microfiber sensor that operates on the principle of self-mixing interference. This sensor is designed to detect respiratory vibration signals emanating from the wrist. The sensor comprises a biconical optical fiber heated, stretched, and enclosed within a polydimethylsiloxane (PDMS) layer. The microfiber serves as a sensor component for detecting respiratory vibrations, while the PDMS film enhances the sensing area to enhance the comfort of wearing. The experimental results show that the sensor has a fast response time (3 ms), good repeatability (>30 000 cycles), and very high sensitivity. It successfully detected vibration signals from the wrist across four distinct respiratory states: apnea, shortness of breath, deep breathing, and normal breathing. We created a dataset from the collected signals to train the 1 dimensional convolutional neural network (1DCNN) model. This model demonstrated intelligent monitoring of respiratory status, as evidenced by tests that yielded up to 98% accuracy. As a result, we successfully implemented intelligent monitoring of respiratory states. This study showcases the effectiveness of using 1-D convolutional neural-network-assisted microfiber sensors for monitoring respiratory status. The findings suggest that these sensors have prospective applications and can provide valuable insights in the healthcare industry.