Automatic arrhythmia analysis techniques and QRS detection provide convenience for the prevention and diagnosis of cardiac disease. The existing studies generally study arrhythmia classification and QRS recognition separately, which requires two different models and may result in time and resource wasting. To realize the goal of arrhythmia classification and R-peak detection of electrocardiograms at the same time, we proposed a method for multi-class arrhythmia classification and R-peak detection method based on one-dimensional U-net with skip-connection and data augmentation. First, the ECG signals were preprocessed by filtering and segmentation, and the ECG annotations were also processed into pixel labels of equal length. Then, we applied data augmentation techniques such as changing gain, adding noise, and flipping signals up and down to increase the diversity of the data. Finally, a modified one-dimensional U-net with skip-connection layers was built to adaptively extract deep features and to detect the arrhythmia type and R peaks of ECG at the same time. We set up an 8-category experiment using five publicly available datasets, and the experimental results show that the macro average F1 value is 94.57%, which is over 4.3% and 3% higher than that of original U-net and the skip-connection U-net without data augmentation, respectively. Meanwhile, the F1 of R-peak detection is 99.64%.