Accurate abnormality detection of dynamic ECG signal is still the diffi culty point in (electrocardiogram,ECG) signal analysis.In this case,the paper designs an optimized Res-Net re sidual network model based on convolutional neural network.Through the classic Res-Net50mode l,the expansion convolution is added into the convolution layer,and the loss function is replaced by the approximate formula of the opposite number of evaluation indexes,so as to o btain higher recognition rate of abnormal ECG features.Several hyper-parameters including l earning rate, kernel size,weight decay,are carefully tested and the optimal values are selec ted as the final parameters.ICBEB ECG database which includes 9kinds of arrhythmias in the 12l ead ECG signal is selected to verify the feasibility.The experimental results show that the classification accuracy of the proposed model reaches 0.95,and the F1score of the model class ification evaluation index is 0.89with lr=0.01,dropout=0.5,weight_decay=0.0001,which shows the potential in clinical diagnosis.