High-frequency surface wave signals in short-duration urban traffic noise often suffer from low signal-to-noise ratio and strong background interference, leading to significantly reduced accuracy and efficiency in dispersion curve extraction. To address this, an intelligent extraction method integrating phase-weighted stacking improved phase-shift method and DBSCAN density clustering is proposed. The method innovatively utilizes instantaneous phase coherence to construct a weight function, effectively suppressing array response sidelobes through a phase-weighting mechanism, significantly enhancing the aggregation and discernibility of dispersion energy under high-frequency and low signal-to-noise ratio conditions. Furthermore, the DBSCAN algorithm is introduced, leveraging its adaptive clustering capability for discrete energy clusters to accurately identify the fundamental dispersion mode, overcoming the limitations of traditional methods that are sensitive to noise and reliant on manual intervention. Numerical simulations demonstrate that the proposed method achieves a deviation of less than 3% from theoretical dispersion curves under low SNR conditions, outperforming the Kmeans method, with an average computational efficiency improvement of approximately 40%. It enables fully automatic extraction of high-frequency dispersion curves with high precision without manual intervention, providing a reliable technical means for rapid detection of urban shallow geological structures.