Super-Resolution Reconstruction of Seismic Images Based on Deep Residual Channel Attention Mechanism

In seismic data interpretation, the resolution and signal-to-noise ratio of seismic profile images directly affect the subsequent interpretation of seismic data. High-frequency information represents details and boundaries, such as stratigraphic interfaces and small-scale geological features. Considering the characteristics of seismic data, this paper adopts a GAN network based on deep residual channel attention mechanism to achieve seismic profile reconstruction and random noise suppression, and improve the resolution of seismic data. The generator network in this algorithm adopts an RIR (Residual in Residual) structure, which uses both long skip connections and short skip connections to bypass abundant low-frequency information and directly learn high-frequency information in seismic data. This structure extracts deeper high-level feature information from seismic data and introduces residual blocks with channel attention (RCAB), allowing network weights to favor channels with more stratigraphic features, thereby improving the reconstruction performance of the generator. Additionally, we use a mixed loss function algorithm to train the network, incorporating a perceptual loss function to further enhance the perceptual quality of the reconstructed images and employing a relativistic adversarial loss (RaGAN) to obtain sharper seismic image edge information. To address the issue of network training relying on a large amount of training data pairs, we synthesize high-resolution seismic data and establish a high-order data degradation model to obtain low-resolution seismic data with different degradation factors. The experimental results of processing synthetic seismic data and actual data show that the proposed workflow can significantly improve the quality perception of the original data, achieve an ideal signal-to-noise ratio, further enhance the resolution effect of the reconstructed profile images, and construct more accurate structural and stratigraphic feature maps than the original seismic images.