Identifying the Respiratory Sound Based on Single-Channel Separation and Hyperdimensional Computing

In intensive care units (ICUs), efficient respiratory management, particularly sputum suction in weakened patients, is critical. Traditional stethoscope-based methods for respiratory sound analysis in tracheal sputum assessment are time-consuming and often struggle to differentiate between cardiac and respiratory sounds, affecting sputum detection accuracy. To address these issues, we propose identifying respiratory sound based on single-channel separation and hyperdimensional computing (IRS-SSHC). Specifically, the proposed method first employs an encoder-decoder framework to effectively separate heart and respiratory sounds in the time domain. Then, it segments respiratory sounds using short-duration energy, where each segment is represented by a 1024-D vector space. Next, it utilizes light gradient boosting machine (LightGBM) based on the vector space for classification. Experimental results show that the classification ACC of IRS-SSHC is 97.9%, which outperforms existing methods.