Aiming at the limitation of the storage capacity and computing power of the body sensor network（BSN）wearable devices

in order to effectively improve the recognition rate and efficiency of the electrocardiograph（ECG）analysis model

a method for extracting ECG fusion features for BSN is proposed.The Pan-Tompkins（PT）algorithm is used to detect the QRS wave group in real time and extract the time-domain features.Then the lightweight sparse auto-encodings（SAEs）network is used to extract the depth features and reduce the dimensions of the continuous ECG signals

and the feature fusion is completed by means of vector span.The results of simulation experiments using ECG data in MIT-BIH database show that the ECG analysis model based on fusion features has high accuracy and less time consumption

and can meet the performance requirements of ECG feature extraction oriented to the body sensor network.