In order to improve the accuracy of modulation recognition, the invention provides a signal modulation recognition algorithm based on data enhancement and a convolutional neural network. The method comprises the following steps: firstly, processing a time domain signal in a data set RML2016.10a into a constellation diagram, and dividing pictures into a training set and a test set according to a proportion of 8: 2; secondly, processing the training sets by using four data enhancement methods of rotation, random erasing, overturning and CutMix to obtain four enhanced training sets; respectively inputting the four types of enhanced training sets into a GoogleNet network for training, and continuously optimizing network parameters according to back propagation to obtain a trained network; and finally, sending the test set into the trained GoogleNet network for testing, drawing a curve of which the correct rate changes along with the signal-to-noise ratio, and comparing the improvement effects of different data enhancement methods on the correct recognition rate to obtain an optimal data enhancement method. Compared with other modulation recognition algorithms without data enhancement, the method has the advantages that the risk of model overfitting is reduced, and the generalization ability of the model is improved.