Recovering reverberation interference striations by a conditional generative adversarial network

In shallow water, the reverberation interference striations (RISs) are often polluted by serious random seafloor scattering, and it is the main limitation for the performance of active sonar. In this letter, a conditional generative adversarial network (CGAN) is used to recover the clear RISs from distorted ones. The experimental distorted RIS data induced by an explosive source are imported into the well-trained CGAN, and the precise interference striations are obtained. This means that the deterministic part of reverberation can be extracted exactly from the stochastic scattering field. The CGAN is robust when reverberation-to-noise ratios are higher than 2 dB.