The construction of mining subsidence basin is of great significance to the study of surface deformation characteristics and failure mechanism. While airborne LiDAR point cloud′s application in constructing subsidence basins remains experimental, this paper proposes a novel approach considering the significant effects of vegetation and topography on mountain subsidence basins. After analyzing the impact of slope variation factors and point cloud vegetation density on basin precision, we introduce a coarse difference filtering method based on a sliding window. This is followed by deploying wavelet theory to filter out high-frequency errors, determining the subsidence basin boundary through sample statistics, and ultimately constructing a highly precise subsidence basin. The results show that the mean square error of the subsidence basin decreases from 0.195 m to 0.072 m, with the maximum error dropping from 1.342 m to 0.245 m. This provides a fresh methodology for constructing subsidence basins in mountainous regions.