Optimization of Ultrasonic Respiratory Signals based on Supervised Learning

Abstract

There are various methods to monitor human respiration. Traditional methods of monitoring the human respiratory process often rely on complex medical equipment, which makes it difficult for users to operate. Nowadays, more and more researchers are focusing on smartphone-based systems that use mobile phones to transmit ultrasound to the chest and abdomen of the human body and use the unique reverse echo of ultrasound to collect respiratory signals. However, this method is easily disturbed by the environment, clothing, equipment, and other factors. Thus, the accuracy is unsatisfactory. This paper presents a method to optimize the respiratory signals collected by ultrasound. This method is based on supervised learning. Piezoelectric sensors and mobile phones are used to monitor human respiratory signals. A Long-Short Term Memory (LSTM) is established to learn the expression from ultrasonic signals to piezoelectric signals to improve the accuracy of signal acquisition. The results show that the model has good performance in both the time and frequency domains, achieving less than 0.05 mean absolute error (MAE) and 0.8779 intersections over union (IoU). The model can be used to optimize the ultrasound respiratory signals.

Publication
In Proceedings of the 28th IEEE International Conference on Parallel and Distributed Systems
Fan DANG
Fan DANG
Research Assistant Professor

My research interests include AIoT, edge computing, and mobile security.