oa Global Information Correlated Multi-Scale Integrated GAN for DAS Seismic Data Denoising

Distributed acoustic sensing (DAS) technology has been gradually applied to vertical seismic profiling (VSP), where the generated DAS VSP seismic data are often severely disturbed by multiple types of complex DAS noise. Therefore, data denoising plays an important role in obtaining high-quality geologic information. In recent years, generative adversarial network (GAN) has been widely used in seismic exploration data denoising. Based on the features of DAS VSP data and the the inspiration from adversarial learning, in this paper, we propose a deep learning network architecture DuGAN to perform multi-scale denoising and global information discrimination to better meet the requirements of high-precision in DAS VSP data denoising. Our method takes GAN as the basic architecture and selects the multiscale codec network U-net to explore the potential correlation of DAS data at different scales and more robust DAS signal feature representation. In addition, to adjust on the situation that the optimization target space in GAN only fucuses on the generative network, DuGAN is more inclined to emphasize the global role of the discriminator, so that the whole network ensures the integrity of the effective signal structure from a global perspective. In order to recover DAS reflection signals more accurately, we use content constraint and adversarial constraint to design a loss function in adversarial training to guide the training of the network, and tilted the objective optimization space towards the discriminator, accomplishing a near-ideal DAS signal recovery in both time and frequency domain. According to the network characteristics of the constructed global multi-scale DAS denoising network, the training set is reasonably adjusted and expanded. While controlling the size of the training samples, the representativeness and typicality of the DAS training set are improved, and a standardized data training set approximating the actual signal is obtained. Experiments on synthetic and field DAS seismic data show that DuGAN has excellent performance in denoising tasks, which not only effectively reduces complex DAS noise but also better preserves effective information, demonstrating its potential capability in oil and gas resource exploration, and provides practical and reliable technical support for intelligent and real-time processing of DAS VSP data.