Seismic denoising is an important processing step before subsequent imaging and interpretation, which consumes a significant amount of time, whether it is for Quality control or for the associated computations. We present results of our work in training convolutional neural networks for denoising seismic data, specifically attenuation of surface related multiples and removal of overlap of shot energies during simultaneous-shooting survey. The proposed methodology is being explored not only for its ability to minimize human involvement but also because of the trained filter's ability to accelerate the process, hence, reduce processing time.

翻译：地震去噪是后续成像与解释前的重要处理步骤，无论是质量控制还是相关计算都需要耗费大量时间。我们展示了在训练卷积神经网络用于地震数据去噪方面的研究成果，特别针对面波多次波的衰减以及同步激采集中炮能量重叠的去除问题。该方法的探索不仅旨在减少人工参与，更因其训练滤波器的能力可加快处理流程，从而缩短处理时间。