<p>High-resolution seismic data with high-fidelity amplitude information is crucial for reservoir prediction and detailed characterization in complex geological settings. Eliminating the effects of stratigraphic absorption and attenuation is a key step in seismic data processing. Conventional time-frequency deconvolution methods, such as those based on the Gabor transform, often lack accuracy and adaptability due to fixed time windows, which is particularly problematic in the heterogeneous basins of the South China Sea. To address these limitations, this study proposes a composite method. First, an improved Generalized S-transform is employed to obtain a high-resolution time-frequency representation of non-stationary seismic data. Subsequently, the wavelet is extracted in the power domain, a key innovation that eliminates the conventional and often unrealistic white-spectrum assumption for the reflection coefficient, thereby significantly enhancing algorithmic stability. Leveraging the flexibility of the Generalized S Transform, we then apply a synchrosqueezing operation to the time-frequency spectrum, effectively improving its concentration and resolution. The combined Power Domain Deconvolution (PDD) and Synchrosqueezed Generalized S Transform (SSGST) methodology is applied to field data from the South China Sea. Results show a clear increase in reservoir-related energy with decreasing frequency. The composite method effectively improves seismic data resolution and provides a reliable tool for thin-layer reservoir identification..</p>