In Low Earth Orbit (LEO) satellite communication systems, the high dynamics of satellite coverage and frequent user handovers pose significant challenges for efficient resource scheduling and signaling overhead management. This paper proposes a Dynamic Spatio-Temporal Feature Fusion User Clustering and Group Handover Strategy (DTSC) based on dynamic spatio-temporal feature fusion, addressing these challenges through multi-dimensional feature modeling and sparse spectral clustering algorithms. The method first filters valid users via elevation angle, predicts remaining handover time using regression analysis, and achieves optimal user-target satellite matching via a weighted sum model. It then introduces a dynamic time window mechanism to quantify handover urgency (time deviation \(\varDelta t\) ), fuses geographical distance to construct spatio-temporal similarity metrics, and reduces computational complexity to O(5N) via a sparse spectral clustering algorithm retaining Top-5 nearest neighbors. The group handover mechanism classifies three priority levels based on \(\varDelta t\) , optimizing handover order through group head signaling aggregation and resource pre-allocation. Simulation results show that compared with traditional single-user strategies, the DTSC strategy reduces signaling overhead by up to 52% in dense scenarios and 35% on average, while improving handover success rate by 28%, providing an efficient solution for dynamic management of large-scale users in LEO satellite networks.