Paper reading and analysis: CorNET Deep Learning Framework for Heart Rate Estimation and Biometric Identification

Consequences of motion artifacts: the largest spectral peak is not heart rate.

Algorithm architecture:

The algorithm is used to solve two problems: 1. Use regression to obtain heart rate; 2. Use classification to identify subjects;

Note: There is a problem of data imbalance in the identification of subjects, because 20 subjects are doing the two-category task: 1:19. The solution is to weight Loss. (The class loss is weighted to offset the class imbalance.)

note: The input of the algorithm is the original PPG signal with length L=1000.

Network Architecture:

Parameters used by the network:

Experimental results:

参考：
CorNET: Deep Learning Framework for PPG-Based Heart Rate Estimation and Biometric Identification in Ambulant Environment