Reliable foundation design for power-transmission towers in mountainous terrain requires accurate knowledge of shallow weathering and bedrock geometry, yet steep relief, irregular bedrock, and uneven weathering limit point-based drilling and conventional surveys at the approximately 22 m × 22 m tower footprint. Here, natural-source seismic frequency imaging was adapted to this setting through a multi-constrained characteristic-frequency workflow, a diagonal-crossing observation system, and terrain-adaptive acquisition. Three survey grids comprising 138 stations were acquired in Minhou, Fujian, using 1 m receiver spacing and 20 min records. The resulting dimensionless normalized relative wave-impedance profiles delineated the silty-clay cover and completely, strongly, and moderately weathered tuff, including a site with 19 m internal relief. Against 13 interfaces from five boreholes, signed depth errors ranged from −0.80 to 1.20 m (root mean square error [RMSE] 0.61 m; mean absolute error [MAE] 0.53 m). Supplementary sensitivity analysis shows that velocity-prior and characteristic-frequency biases can produce depth errors comparable to the field residuals, whereas the chosen 1 m spacing contributes relatively small geometric interpolation error for interfaces varying over wavelengths of several meters. A geometry-only 3D sampling test further shows that the diagonal layout can miss small off-line bodies and therefore cannot replace true 3D investigation. The method is consequently best used as a borehole-calibrated, continuous-coverage complement for shallow foundation characterization rather than as a stand-alone lithology classifier.