With the development of digital techniques, cardiovascular disease diagnosis techniques based on electrocardiogram (ECG) signals has made significant progress. However, there are currently a large number of physical (paper) ECG in the whole world, which usually contains a huge of rich information of cardiac care hasn't been utilized in digital ECG-based diagnosis approaches. The effectively digitization of physical ECG might be the key to improving the accuracy of cardiovascular disease diagnosis. Focus on the digitization of physical ECG, this paper proposes an ECG digitization method based on ResUNet and YOLO-V8 algorithm. Specifically, the ResUNet is used to remove the background grid and noise from the ECG image, and then the YOLO-V8 algorithm is applied to segment the regions of interest within the ECG signals. The ECG of each lead is extracted based on pixel points and waveform reconstruction is performed at last. For ECG images abnormal detection, this paper proposes a MB_SeResnet (Multi-branch Squeeze-and-Excitation Residual Network) model, which is designed with three branch structures that utilize convolution kernels of different sizes to extract feature information at various scales. By incorporating the Se module, the model can more effectively emphasize important features, thereby improving the accuracy of classification. The method achieves an F1 score of 0.683 on the test set. In the unofficial stage classification task, our team (Proton) ranked 11th.