Investigation of the electrophysiological mechanisms that induce arrhythmias is one of the most important issues in scientific research. Since computational cardiology allows the systematic dissection of causal mechanisms of observed effects, simulations based on the ionic channel mathematical models have become one of the most widely used methods. To reduce themanual classification of different types of membrane potential patterns produced during simulations, a convolutional neural network is developed in this paper. The model includes 4 convolution layers, 4 pooling layers and a fully connected layer. An activation function of ReLU is used. Before machine learning, all the pattems are calibrated, cut, and normalized to a uniform format with a size of 256 ×256 . The contour boundary of each pattern is extracted using the maximum between-class variance method. In the examination, the proposed learning algorithm shows a recognition accuracy of 97% on test data set after training.