SVM-Based Spectral Analysis for Heart Rate from Multi-Channel WPPG Sensor Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Wrist-type photoplethysmography (WPPG) offers convenient heart rate monitoring.
• Motion artifacts from hand movements significantly contaminate WPPG signals, challenging accurate heart rate estimation during physical activity.
• The increasing popularity of wrist-worn devices necessitates robust WPPG signal processing techniques.

Purpose of the Study:

• To propose a mixed approach (Mix-SVM) for accurate heart rate estimation using multi-channel WPPG and acceleration signals.
• To effectively remove motion artifacts from WPPG signals during intense physical exercise.
• To validate the performance of the proposed method using a public PPG database.

Main Methods:

• A combination of Principle Component Analysis (PCA) and adaptive filtering is used to reduce initial motion artifacts.
• Motion artifact removal is further addressed as a sparse signal reconstruction problem, leveraging the correlation with acceleration signals.
• A spectrum subtraction method is employed for effective artifact elimination, followed by SVM-based spectral analysis to identify heart rate peaks.

Main Results:

• The Mix-SVM approach demonstrated a low average absolute error of 1.01 beats per minute.
• A high Pearson correlation coefficient of 0.9972 was achieved, indicating strong agreement with ground truth.
• Experimental results confirm the method's efficacy in multi-channel WPPG-based heart rate estimation.

Conclusions:

• The proposed Mix-SVM method significantly improves the accuracy of heart rate estimation from WPPG signals.
• This approach is robust in the presence of strong motion artifacts encountered during intense physical exercise.
• Mix-SVM shows considerable potential for reliable heart rate monitoring in wrist-worn wearable devices.