Underwater target perception algorithm based on pressure sequence generative adversarial network

In recent years, underwater target perception algorithms based on deep learning have attracted extensive attention. Deep learning can independently learn to extract features from a large number of labelled data, which improves the robustness of underwater target perception accuracy. However, owing to the high collection cost in the underwater natural environment and the high time cost of simulated calculations, it is often unrealistic to provide a large number of labelled data. To solve this problem, this paper proposes an underwater target perception algorithm based on a generative adversarial network (GAN). This GAN uses the transformer model to augment the samples of a small number of simulated underwater pressure sequences and then establishes a multilayer gated recurrent unit (GRU) network to recognise the azimuth, distance, and velocity of underwater targets. The experimental results show that the method proposed in this paper can effectively realise underwater target perception, and with an accuracy of 97.86%, and the root mean square errors of the target distance, azimuth, and velocity estimations are 0.1244, 0.9828, and 0.8271, respectively.