Contactless Respiratory Monitoring Using Acoustic Convolutional Neural Network Classification

Continuous respiratory monitoring plays a crucial role in both clinical and non-clinical applications, providing valuable insights into physiological health. This paper presents a sustainable, contactless respiratory monitoring framework that integrates acoustic sensing with a lightweight convolutional neural network (CNN) optimized for low-power embedded platforms. Breathing sounds are processed using wavelet-based denoising and Mel-Frequency Cepstral Coefficient (MFCC) extraction, achieving 94.8% classification accuracy with an inference latency of 0.3 s per frame. The quantized model deployed on a Sony Spresense microcontroller reduces memory usage by over 90%. By eliminating disposable sensors and minimizing energy consumption, the proposed approach delivers an eco-efficient, scalable, and real-time solution for continuous respiratory assessment.