Abstract: Surface wave analysis, used to characterize near-surface shear-wave velocity structures, has been widely applied to various distinct fields. Extracting surface wave dispersion information serves as a significant step in surface wave analysis. Conventional methods for picking dispersion curves primarily include manual picking and semi-automatic picking in manually delineated dispersion energy zones. Given the large volume of seismic data used in oil and gas exploration, manual processing methods struggle to satisfy the requirements of practical production. Therefore, this study proposes a method for automatically picking multi-mode surface-wave dispersion curves based on an unsupervised machine learning approach. First, some background noise points in a dispersion spectrum are removed using energy threshold filtering. Second, different surface wave modes in the dispersion spectrum are determined using the energy density-based spatial clustering of applications with noise (E-DBSCAN) algorithm. Third, energy peaks in zones corresponding to different surface wave modes are searched, followed by noise elimination through Kalman filtering and smoothing. As a result, multi-mode dispersion curves are obtained. The proposed method incorporates dispersion energy values, offering high reliability. Numerical experiments demonstrate that the automatically picked dispersion curves accurately match the theoretical dispersion curves. Meanwhile, method comparison and field data tests further confirm the reliability of the proposed method.