A Photoplethysmography Signal Compression Method of Extreme Learning Machine
Yan Zhou, Aimin An, Jicheng Liu, Yongxin Chou · 2023
As one of the important physiological signals for human health monitoring, there are massive photoplethysmography (PPG) signals and excessive demands on storage and network communication resources for long time monitoring. Therefore, it is desirable to compress the PPG signal where redundancy exists without losing important information. In this paper, a PPG signal compression method with extreme learning machine (ELM) is proposed. ELM does not require iterative adjustment of the hidden layer parameters and has the advantages of high generalization ability and fast learning speed. First, the PPG signal is segmented and processed as input and output data. Second, the parameters of the hidden layer based on the set training set is calculated. Then, the obtained parameters are used to compress and reconstruct the PPG signal, one PPG wave band at a time, until the end of the training. Comparing with the error Back Propagation Training (BP) neural network, for the 4 min sampled data, the average time to compress and reconstruct the PPG signal is 0.026±0.007, the MSE is 0.004±0.004 and the MAD is 0.031±0.026.