In wireless communication，modulation classification is an important part of non-cooperative communication，and it is difficult to classify various modulation schemes efficiently using conventional methods with both high recognition accuracy and low complexity. Deep learning method is used to deal with this problem，and good results can be achieved. In the underwater acoustic communication environment，due to the particularity of the communication environment，the modulation classification of signals is more difficult in here than the terrestrial communication. This paper innovatively adopts the modified residual structure form of the deep learning method to distinguish various commonly used modulation schemes in various underwater acoustic communications. By reasonably selecting the super parameters of the deep residual network，the over-fitting problem is effectively overcome，and the good recognition effect is achieved.
