Abstract: The Qiongdongnan Basin, situated in the northern part of the South China Sea, is a geologically significant and resource-rich marine area, characterized by extremely complex seabed topography that transitions seamlessly across shallow coastal waters, gently sloping continental shelves, steep shelf break zones, and abyssal deep-sea basins. Within this basin, the vicinities of the rugged, uneven seabed—marked by abrupt topographic variations such as submerged ridges, fault scarps, and scattered seamounts—have emerged as key exploration zones, as geological surveys and preliminary prospecting have identified a multitude of potential hydrocarbon and mineral reserve targets in these sub-seabed formations. However, the amplitude and frequency of seismic data under the rugged seabed in deep water are distorted, and there is a large gap between the seismic data collected on land and that under the rugged seabed. Traditional signal compensation methods fail to effectively process such data. In order to obtain high-quality seismic data and accurately evaluate targets under the rugged seabed, it is urgent to develop a reasonable and effective compensation method. Inspired by time-frequency analysis and compressive sensing methods, this paper designs a compressive sensing compensation framework based on W transform. By constructing the compensation matrix between the reference channel and the target channel, the amplitude compensation of seismic data can be quickly realized in one step. The sparse transform regularization method is introduced to ensure the high signal-to-noise ratio of the compensation data and improve the quality of the seismic data. This method is used to compensate the seismic data of the northern part of the South China Sea, and the consistency of the energy distribution of the seismic data profile after compensation has been improved. The successful application of the signal compensation processing proves the feasibility of the method and can provide a reference for the signal compensation processing of similar seismic data.