In order to achieve non-invasive detection of human hemoglobin concentration

a prediction model for hemoglobin concentration based on BP neural network is established. MAX30102 sensor and the chip STM32 F103 are used as data acquisition systems to collect the photoplethysmographic pulse wave data of the human fingertip.The baseline drift noise in the original pulse wave signal is eliminated by wavelet transform algorithm

and 22 PPG time domain feature parameters and 3 human body feature parameters are extracted. Combined with Relief-PCA algorithm

these 25 characteristic parameters were screened

and 10 key feature parameters are determined as input values to construct BP neural network model. A total of 115 sets of data are collected

80 sets are randomly selected as the training set of the model

and 35 sets are the test set. The reference value is the minimally invasive value measured by the Misson HB hemoglobin analyzer. After comparative study

the correlation coefficient reaches0. 88 and the mean square error is 0. 71 g/d L. Studies show that the hemoglobin concentration detection system based on BP neural network can better measure the hemoglobin concentration.