IoT environmental sensing for home-based rehabilitation in COPD patients

Home-based rehabilitation management is crucial for chronic obstructive pulmonary disease. However, existing monitoring technologies face challenges such as poor compliance with contact-based devices, scarcity of pathological samples, and inadequate long-term modeling capabilities. This paper proposes a respiratory monitoring system based on Internet of Things environment perception. This system utilizes commercial Wi-Fi channel state information to achieve non-contact respiratory monitoring and combines an environmental sensor network. The experiments are based on 90-day real home Channel State Information (CSI) data collected by 32 community-dwelling elderly individuals and 180-day environmental and health data of 50 Chronic Obstructive Pulmonary Disease (COPD) patients from the public dataset. The main innovation of this paper lies in: deeply integrating environmental parameters with CSI respiratory signals; proposing a distributed matching-enhanced SiamGAN, using the maximum mean discrepancy loss and diversity regularization term, to expand the sample size of acute exacerbation by 3.5 times and increase sample diversity by 29.5%; constructing a time–frequency attention Transformer model TCFormer to enhance the representation of respiratory features in long sequences. The experimental results show that the average absolute error of respiratory rate monitoring is 1.13 times/minute, which is 8.9% lower than the optimal deep learning method; the recognition accuracy of COPD acute exacerbation binary classification is 92.2%, and the Macro F1 is 91.8%; the inference delay of edge devices is 9.3 ms. This system has achieved high-precision and low-latency COPD home-based rehabilitation monitoring in real-life settings.