Aiming at the anomaly identification problem of electro cardiograph（ ECG） signals

an improved Cifar network model based on deep learning is proposed

using this model to extract ECG features automatically

the extracted features are fed into the Logistic classifier for classification. the scaling transform is used to obtain the multi-resolution representation of the image

this expression is consistent with timing characteristics of the ECG

overcome the problem of network overfitting caused by small amount of data. The ECG sensor is used to collect ECG signals of volunteer

and a training test set is established combined with the normal sinus rhythm database（ NSRDB） and the sudden cardiac death database（ SDDB）

which is used to prove the validity of the method. In the experiment of positive-abnormal ECG classification

the improved cifar network model can achieve accuracy of92 %

sensitivity reaches 94 %

AUC value is 0. 969

which is better than traditional scale invariant feature transform（ SIFT） + support vector machine（ SVM） algorithm and back propogation（ BP） algorithm

obtain a better auxiliary diagnostic effect.