The invention relates to an electrocardiogram classification method based on a lightweight neural network, in particular to a method for realizing automatic classification of arrhythmia through single-lead electrocardiogram data and a lightweight neural network model. The method comprises: firstly, a limb II lead electrocardiosignal containing enough information; secondly, compressing the feature dimension by using a convolutional layer with a convolution kernel size of 1 and a global average pooling layer; and finally, rapidly extracting features by using streamline depth separable convolution. By using the single-lead data set and the lightweight neural network model, the model operation speed can be greatly improved on the basis of ensuring the model accuracy.