Paper reading and analysis: PPGnet: Deep Network for Device Independent Heart Rate Estimation from Photoplethysmogram

main content:

Propose a neural network architecture for heart rate estimation from PPG signals;

Network Architecture:

Network design:

1. Use 8s PPG signal, truncate 8 parts, each 1s, spliced ​​as input, PPG sampling frequency is 125Hz;

2. Use $1\times 5,1\times 20,1\times 40,1\times 60,1\times The\; 80$ % convolution kernel is convoluted and spliced ​​separately to provide the characteristics of different receptive fields;

3. Followed by 8 convolution blocks, the convolution block contains two sub-blocks, each sub-block has convolution, batch normalization and RELU;

4. Use $8\times The\; LSTM\; of\; 125$ operates on the basis of 1;

5. Combine the results of 3 and 4;

6. Transfer the spliced ​​matrix of 5 to $384\times 125$ LSTM modules;

7. Finally connect to the fully connected layer;

result:

参考：
PPGnet: Deep Network for Device Independent Heart Rate Estimation from Photoplethysmogram