Full-waveform inverse problem is a significant research field, embedded with problems and solutions from different studies. The boost of deep learning and neural network lead to the occurrence of more and more data-driven solutions, therefore, the study in this field is more specifically divided.
In this paper, the regularization part of the full-waveform inverse problem is re- searched by using a generative adversarial network(GAN). From building of the sce- nario, mimicking the defect problem set, to the training process of the GAN and the evaluation of the formed regularizer. To conclude, the paper draws the full method- ology to deal with problem and the possibility of the regularization through a GAN trained discriminator is validated. The regularizer is evaluated as valid in distinguishing between real images and counter facts.