Optimization of migrated images based on cycle generative adversarial network
This paper presents an optimization method for reverse time migration(RTM)imaging using a cycle generative ad-versarial network(CycleGAN).Based on the conventional RTM method,the CycleGAN framework is introduced,compri-sing two generators and two discriminators.To prevent overfitting,an identity loss function is added alongside the adversarial and cyclic consistency loss function.A training dataset is constructed to train the network,enabling it to learn the mapping between the conventional RTM imaging results and least-squares RTM imaging results.Finally,synthetic data and real data are used to test the network̍s performance.The prediction results from both datasets demonstrate that the proposed method ef-fectively enhances computational efficiency while producing high-precision and high-SNR imaging results.
CycleGANresidual netinverse Hessianleast-squares reverse time migration
万方数据
维普
