Abstract: The microtremor surface waves collected from karst areas, where near-surface conditions are complex, often show low signal-to-noise ratios (SNRs) and poor data quality, thus restricting the detection accuracy. Hence, this study proposed a microtremor surface wave imaging method based on singular value decomposition (SVD). First, the single-trace virtual-source empirical Green's function obtained by cross-correlation was reconstructed through SVD. The signal components corresponding to large singular values were extracted to reconstruct the wave field, thereby effectively suppressing the noise and improving the SNRs of surface waves. Second, the virtual-source multi-trace records were constructed using the reconstructed signal with high SNRs. The dispersion energy spectrum was derived through the F-K transform, obtaining significantly improved surface wave dispersion bands in terms of resolution and continuity. Third, the frequency dispersion curves were extracted from the optimized dispersion energy spectrum. The joint inversion strategy employing the genetic algorithm and damped least squares was used to further improve the imaging accuracy, considering the globally optimal solution and convergence efficiency. Fourth, by integrating the inversion results of all survey points in the study area, a two-dimensional shear wave velocity profile was constructed to reveal the underground structural characteristics and the spatial distribution of karsts in the study area. The practical application in a karst area demonstrates that the proposed method can effectively improve the imaging quality of the karst area with complex near-surface conditions, providing reliable technical support for the fine-scale structure detection and geological interpretation in karst areas.