Based on the formation mechanism and distribution characteristics of the pulse wave characteristics， this paper proposes to use PDMS packaged fiber grating flexible sensor to detect the human wrist pulse wave signal， aiming at the four most common types of pulse waves： obvious， hidden and partly obvious. The pulse wave signal feature extraction method based on the time-domain differential period ratio uses the relative position and proportional relationship of each feature point in the pulse wave time-domain differential signal as feature parameters， and realizes a comprehensive algorithm from pulse wave detection to feature point extraction. The results show that for the 4 050 pieces of experimental data collected， the algorithm can accurately identify the characteristic points of the starting point and the crest. In the resting state， the identification accuracy of the tidal wave d and e points is 98.28% and 97.25%. The recognition accuracy rates of points f and g are 98.14% and 99.19%； in the state of exercise， the recognition accuracy rates of tidal waves d and e are 94.23% and 90.77%， respectively， and the recognition accuracy rates of dicrotic waves f and g are 91.93% and 95.38% respectively.