基于DCGAN的水声数据仿真方法

A Simulation Method of Underwater Acoustic Data Based on DCGAN

研究基于深度卷积生成-对抗网络(Deep Convolution Generation-Antagonism Network,DCGAN)实现水声数据仿真的方法,通过设计适配的DCGAN模型,迭代训练出能够拟合随机高斯噪声与水声小波图像之间映射关系的网络模型参数,实现水声数据的仿真。由提前训练好的判别网络验证仿真数据的可靠性,验证结果证明,提出的方法可作为一种智能化水声数据模拟器的实现方案。

This paper has carried out the research on the simulation method of underwater acoustic data based on the deep convolution generation-antagonism network(DCGAN).By designing an adapted DCGAN model and iterative training the network model parameters which can fit the mapping relationship between random Gaussian noise and underwater acoustic wavelet image,to realize the simulation of underwater acoustic data.The reliability of the simulation data is judged by a pre-trained discriminant network.It is proved that this method can be used as an intelligent underwater acoustic data simulator.