A CNN and GRU combined ECG signal classification method based on the self-encoding mode is provided. The most representative features of an original signal is extracted to conduct feature extraction by using CNN+GRU, and a lot of training space is saved. On one hand, the adopted GRU (Gated Recurrent Unit) solves the problem of gradient disappearance and gradient explosion during RNN training; on the other hand, the GRU has one gate less than LSTM, the calculation is easier, and the training efficiency can be improved. The GRU has the advantage that over-fitting can be prevented when training samples are less, and more training time can be saved when the training samples are more, and the learning efficiency of the network and the ECG signal recognition accuracy can be improved.