The deep Longmaxi Formation in the southern Sichuan Basin exhibits intricate fracture networks resulting from multiple tectonic episodes, rendering high-resolution fracture characterization crucial for effective deep shale gas exploration and production. Current approaches to fracture prediction in deep shale face two fundamental challenges: (i) the deep target formation necessitates high-fidelity velocity models, yet conventional seismic imaging fails to meet the stringent requirements for refined fracture delineation; (ii) existing prediction methodologies predominantly employ conventional seismic attributes such as coherence and curvature. Single-scale curvature analysis proves inadequate for multi-scale fracture characterization, while curvature attributes remain highly susceptible to noise contamination. To address these limitations, this study first introduces an integrated high-precision velocity modeling workflow incorporating geological constraints, grid tomography, and full-waveform inversion (FWI), representing the inaugural application of FWI to deep shale gas exploration in the Sichuan Basin. Second, based on scale-space analysis theory and seismic curvature analysis, this article proposes a method using a Gaussian filter calibrated based on actual drilling breakpoints to characterize seismic fractures at different scales. Using the high-precision imaging results as input, the proposed method predicts fractures at multiple scales. Field trials in a deep shale gas block of southern Sichuan demonstrate: (i) substantial improvements in imaging accuracy and subtle structure delineation compared to conventional processing; and (ii) superior fracture prediction capabilities relative to conventional curvature-based methods, with enhanced resolution of multi-scale fracture systems, thereby providing critical technical support for deep shale gas development.