Schematic diagrams of seismic random noise attenuation based on the (a) typical supervised learning and (b) Noise2Noise unsupervised learning.|@|~(^,^)~|@|Network architecture of Noise2Noise used in this study.|@|~(^,^)~|@|Network architecture of DnCNN used in this study.|@|~(^,^)~|@|Poststack time migration cube of the Youngil Bay 3D high-resolution seismic data.|@|~(^,^)~|@|The 89^th inline section of poststack time migration data before denoising.|@|~(^,^)~|@|Denoised result and estimated noise of FX deconvolution ((a) and (b)), DSMF ((c) and (d)), Noise2Noise ((e) and (f)) and DnCNN ((g) and (h)).|@|~(^,^)~|@|The energy attribute result (a) before noise attenuation and after applying (b) FX deconvolution, (c) DSMF, (d) Noise2Noise, (e) DnCNN.|@|~(^,^)~|@|The sweetness attribute result (a) before noise attenuation and after applying (b) FX deconvolution, (c) DSMF, (d) Noise2Noise, (e) DnCNN.|@|~(^,^)~|@|The similarity attribute result (a) before noise attenuation and after applying (b) FX deconvolution, (c) DSMF, (d) Noise2Noise, (e) DnCNN.|@|~(^,^)~|@|The frequency spectrum of the 89^th inline poststack seismic section before and after noise attenuation.

© Econ. Environ. Geol.