Abstract: The accurate detection of near-surface low-velocity anomalies (such as cavities), which has always been a valuable and challenging research topic in the field of geophysics, holds great significance for the fine-scale near-surface modeling in urban disaster assessment and the seismic exploration of coalfields under complex conditions. The full-waveform inversion (FWI) of surface waves is suitable for high-precision near-surface modeling. However, there still exist many urgent problems with the modeling in practical applications. To solve the key issues of the FWI of surface waves, such as multi-parameter crosstalk, actual data preprocessing, and source wavelet estimation, this study developed a complete method of multi-parameter FWI of Rayleigh waves to achieve the accurate detection of near-surface cavities. In this method, (1) the S- and P-wave velocities and density of the models were updated synchronously in the process of inversion, thus reducing the negative effects of the deviations of the P-wave velocity and density from their actual values on the accuracy of inverted S-wave velocities; (2) the quasi-Hessian operator constructed using the adjoint state method was employed to conduct gradient preprocessing in order to suppress surface artifacts, enhance wavefield illumination, and improve the characterization ability for small-scale anomalies; (3) to transform the 3D wave field into a 2D wave field, the convolutional factor was used to eliminate the dimension difference between the wavefield forward modeling and the actual data acquisition; (4) to reduce the influence of specific inaccurate parameter models, the corrected filtering method was used to dynamically estimate the source wavelets during the iterative process; (5) to improve the stability of the inversion, a multi-scale inversion strategy was adopted to alleviate the non-convexity of the objective function caused by low-velocity anomalies. The synthetic data and the test results of actual cases show that the models of the S- and P-wave velocities and the density developed using the multi-parameter FWI method of Rayleigh waves were roughly consistent, with the S-wave velocity model being the most accurate. The S-wave velocity model obtained through the inversion of measured data revealed a 4 m × 3 m artificial cavity, whose location and size are consistent with the actual situation. This result demonstrates the method proposed in this study features feasibility and effectiveness in the detection of near-surface cavities.