In this paper, a new lightweight and efficient time-series backbone structure–temporal attention splitting network (TAS) is built, and good estimation of Cuff-Less blood pressure and non-invasive blood glucose is achieved on this model using photoplethysmography(PPG) signals. Blood pressure and blood glucose are two critical indicators of human health, reflecting people’s health well. Convenient and accurate blood pressure and blood glucose measurements are increasingly popular. In this paper, the estimation of non-invasive blood glucose and CuffLess blood pressure are achieved by collecting the PPG signals of 20 volunteers and the PPG signals of 200 patients obtained from the MIMIC-II database, respectively. The experimental results show: on the blood glucose dataset, the TAS model proposed in this paper achieved a standard error of prediction SEP =10.90mg/dL, Pearson correlation coefficient R=0.93, and determination coefficient R 2 =0.86, and 93.94% of the prediction points are located in A when the clinical safety of the model effect is evaluated with a Clark error grid region. On the blood pressure dataset, the absolute mean error MAE =1.48 mmHg and standard deviation STD =2.31 mmHg for diastolic blood pressure (DBP) and MAE =2.85 mmHg and STD =4.65 mmHg for systolic blood pressure (SBP). According to the British Hypertension Society (BHS) treatment protocol, systolic and diastolic blood pressure achieved grade A. These results show that TAS is an accurate estimator that can learn PPG well with blood glucose and blood pressure information.