Electrocardiogram (ECG) detection is the most commonly used diagnostic method of heart disease. However, in the process of ECG signal acquisition, it is often disturbed by noise, which greatly affects the accuracy of ECG signal classification and diagnosis. In order to improve the accuracy and anti noise ability of classification diagnosis, this paper improves and designs an ECG classification and diagnosis model which use deep residual shrinkage network (DRSN) to resist noise automatically and integrate spatial information by global average pooling (GAP). The classification performance of the model is verified on MIT-BIH arrhythmia data set, and its anti noise performance is analyzed and compared with the ordinary convolutional neural network (CNN) model. The experimental results show that the classification accuracy of the designed DRSN + GAP diagnostic model based on AAMI standard is up to 99.3%, and its anti noise performance is better than ordinary CNN model for power frequency and Gaussian noise with different intensity.