Intelligent Ghost Wave Suppression Based on Quadratic Convolution Kernel and Attention Mechanism

In offshore streamer seismic data processing, ghost wave suppression is a crucial step, as it not only enhances data resolution but also restores the signal’s frequency bandwidth. Recent advancements in artificial intelligence have demonstrated significant progress in intelligent suppression technologies for ghost waves within geophysical exploration. By introducing a quadratic convolution kernel, we propose the novel quadratic convolutional attention-U-Net (QAU-Net) for ghost wave suppression, aiming to enhance the network's fitting accuracy. In addition, we establish a curve function that simulates the shape changes of the streamer in the actual acquisition environment. The training dataset exhibits higher realism and more abundant ghost waves generated by shape-changing streamers. In the experimental validation phase, the proposed quadratic convolutional network demonstrates accelerated convergence rates and enhanced function approximation capabilities relative to conventional linear counterparts, as indicated by multiple evaluation metrics. This enhancement stems from the quadratic operations' inherent nonlinear modeling superiority, which facilitates precise approximation of objective functions while maintaining the network's structural configuration. Compared with the conventional horizontal streamer training dataset, the QAU-Net trained with shape-changing streamer dataset is more inclusive to the shape changes of streamer. The proposed method is also successfully applied to field data from the South China Sea, demonstrating that the solution has strong generalization capabilities.