Preprocessing of breathing signals using band-pass filtering and zero-crossing techniques for 1D CNN activity classification

초록

This study proposes a problem-specific preprocessing framework that integrates a band-pass filter (BPF) and zero-crossing analysis to reduce viscoelastic distortion and peak-detection errors in respiratory signals acquired from textile-based capacitive wearable sensors. The proposed framework improved respiratory cycle representation by reducing excessive peak detections and stabilizing cycle segmentation under different motion conditions. The preprocessed signals were subsequently evaluated using a one-dimensional convolutional neural network (1D CNN) for three-class motion-state classification (standing, walking, and running), achieving an accuracy of 94.44%. These results suggest the feasibility of the proposed preprocessing framework for respiratory signal analysis using viscoelastic material-based wearable sensors under controlled motion conditions, while further validation with multi-subject datasets and reference respiratory measurements is required.

키워드