In the study area, southeast of Ordos Basin in China, thick lacustrine shale/mudstone strata have been developed in the Triassic Yanchang Formation. Aiming to study these source/reservoir rocks, a 3D full-azimuth, high-density seismic survey was acquired. However, the surface in this region is covered by a thick loess layer, leading to seismic challenges such as complicated interferences and serious absorption of high frequencies. Despite a specially targeted seismic processing workflow, the prestack Kirchhoff time-migrated seismic data were still contaminated by severe noise, hindering seismic inversion and geologic interpretation. By taking account of the particular data quality and noise characteristics, we have developed a cascade workflow including three major methods to condition the poststack 3D seismic data. First, we removed the sticky coherent noise by a local pseudo -- Cadzow filtering. Then, we diminished the random noise by a structure-oriented filtering. Finally, we extended the frequency bandwidth with a spectral-balancing method based on the continuous wavelet transform. The data quality was improved after each of these steps through the proposed workflow. Compared with the original data, the conditioned final data show improved interpretability of the shale targets through geometric attribute analysis and depositional interpretation.