In light of the escalating global mortality rate attributed to cardiovascular diseases, wearable devices designed for real – time tracking of human physiological data are making rapid inroads. Acoustic sensors, thanks to their low energy consumption, compact dimensions, and cost – effectiveness, are extensively employed in the detection of human physiological states.To begin with, we leverage a convolutional neural network to analyze the morphological characteristics of electrocardiogram (ECG) signals. By using a bidirectional long – short – term memory network, we capture the contextual relationships within these features and develop a CNN – BLSTM network model. Next, we carry out label marking, derived signal extraction, and noise elimination on photoplethysmogram (PPG) signals. We further refine the CNN – BiLSTM model by incorporating an attention mechanism module to achieve blood pressure detection.The experimental findings indicate that the CNN – BiLSTM network model boosts the classification accuracy in class S and class F by 2.02% and 12.94% respectively. Additionally, the recall rate improves by 12.94% and 4.11% respectively, fulfilling the criteria for arrhythmia detection. On the dataset, the optimized CNN – BiLSTM blood pressure prediction model attains a systolic mean error (ME) of 0.8429 mmHg, a mean absolute error (MAE) of 4.5916 mmHg, and a root – mean – square error (RMSE) of 7.1219 mmHg. For diastolic blood pressure, the predicted ME is 2.2577 mmHg and the MAE is 3.0081 mmHg.Lastly, a wearable system capable of simultaneously monitoring blood pressure and atrial fibrillation is developed. This system offers technical backing and practical strategies for the application of wearable medical devices in real – time cardiovascular health monitoring situations.
