Dual-Sensor Signals Based Exact Gaussian Process-Assisted Hybrid Feature Extraction and Weighted Feature Fusion for Respiratory Rate and Uncertainty Estimations

Figures

• 

  Figure 1. Block diagram of the proposed methodology using exact Gaussian process regression (EGPR) with hybrid feature extraction, feature fusion, and weighted features selection using the RNCA model.

• 

  Figure 2. An example of preprocessing, where (a) denotes a PPG signal after removing a high-frequency (HF) signal; (b) denotes a partial PPG signal of 50 s shows maximum points in the ppg signal of about 480 s; (c) is maximum points in the ppg signal (480 s); (d) denotes a resample PPG signal.

• 

  Figure 3. Hybrid feature extraction and weighted features fusion process, where (a) denotes the detailed process of parallel feature extraction using multi-phases (MF) model and (b) denotes power spectral (PS) feature extraction, where RWS denotes a segmented resample wave signal, AC is an autocorrelation for each record or subject.

• 

  Figure 4. High weight features selection is performed using the RNCA model; where (a) is the weighted features and (b) is the regularization parameters λ.

• 

  Figure 5. Top panel (a) represents the MAE and SD compared to the reference RR method acquired from the ECG-based SVR-MF, ECG-based SVR-PS, ECG-based LSTM-MF, and ECG-based LSTM-PS models. Bottom panel (b) shows the MAE and SD compared to the reference RR method obtained from the ECG-based GBA-MF, ECG-based GBA-PS, ECG-based EGPR-MF, and ECG-based EGPR-PS.

• 

  Figure 6. Top panel (a) denotes the MAE and SD compared to the reference RR method obtained from the PPG-based SVR-MF, PPG-based SVR-PS, PPG-based LSTM-MF, and PPG-based LSTM-PS models. Bottom panel (b) shows the MAE and SD compared to the reference RR method obtained from the PPG-based GBA-MF, PPG-based GBA-PS, PPG-based EGPR-MF, and PPG-based EGPR-PS.

• 

  Figure 7. Top panel (a) denotes the MAE and SD compared to the reference RR method obtained using the dual signals based hybrid feature extraction and arithmetic fusion (AF) models., and hybrid feature extraction and feature fusion (FF) models. Bottom panel (b) shows the MAE and SD compared to the reference RR method obtained using the dual signals based hybrid feature extraction and feature fusion (FF) models.

• 

  Figure 8. Top panel (a) denotes the CI estimation using the ECG-based EGPR-MF model to represent the uncertainty for RR estimation; bottom panel (b) is the CI estimation using the EGPR-FF model to represent the uncertainty for RR estimation.