Abstract: Because of the complex marine ambient noise and low signal to noise ratio of the underwater targets to be identified, it is difficult to extract target features, and the target recognition rate is low. Aiming at this problem, a deep learning underwater target recognition method based on improved wavelet threshold is proposed in this paper. In this method, a new wavelet threshold function based on the traditional wavelet threshold denoising method is adopted. The specific threshold relates to the decomposition scale, so as to reduce the background noise and improve the recognition rate of underwater targets. This method performs the wavelet decomposition of the measured ship radiated noise signal and extracts the high-frequency wavelet coefficients of each layer for processing. The time-frequency characteristics of the processed signal are extracted and input it into the subsequent deep learning neural network. The experimental results find that, for the original data set, the deep learning underwater target recognition method based on the improved wavelet threshold can bring the recognition rate of convolutional neural network (CNN) to 88.56%. Further analysis shows that by using a generative adversarial network (GAN), the data samples can be expanded to reach a recognition rate of 96.673%.