Towards Wearable Respiration Monitoring: 1D-CRNN-Based Breathing Detection in Smart Textiles

Monitoring respiratory activity is a key indicator of physiological health and an essential component in smart textile systems for unobtrusive vital sign assessment. In this work, we present a one-dimensional convolutional recurrent neural network (1D-CRNN) for automatic classification of breathing activity from inertial data acquired by a smart e-textile of 59 subjects. The proposed method integrates convolutional layers for local feature extraction with recurrent layers for temporal context modeling, enabling robust segmentation of breathing and noise segments. The model was trained and evaluated using a stratified five-fold cross-validation scheme to account for inter-subject variability and class imbalance. Across different window sizes, the classifier achieved a mean accuracy of 0.88 and an F1-score of 0.92 at a window size of 2000 samples. The best-performing configuration for a single fold, reached an accuracy of 0.995 and an F1-score of 0.99. Furthermore, near-real-time feasibility was demonstrated, with a total processing time-including data loading, classification, segmentation, and visualization-of only 1.76 s for a 250 s measurement, corresponding to more than 100× faster than the recording time. These results indicate that the proposed approach is highly suitable for embedded, on-device inference within wearable systems.