Existing airborne data single-tree segmentation methods exhibit low universality for different forest types, particularly in areas with high canopy closure where the extraction accuracy is notably compromised. Spectral images and LiDAR data from the tropical broad-leaved forest region within the jurisdiction of Haikou City, Hainan Province, China, were employed. Initially, a distance threshold-based single-tree segmentation method was employed to extract tree crown edges from the high-resolution spectral image. Subsequently, the obtained positions of initial detected tree vertices were constrained using the segmented tree crown edges, and precise positioning of single-tree vertices was achieved. Following this, a seed-point-based single-tree segmentation method was applied for final tree extraction in the broad-leaved forest. The results indicated that compared with existing single-tree segmentation methods based on the relative distances between trees, by selecting the optimal segmentation scale in combination with spectral imagery for precise positioning, the issue of over-segmentation caused by traditional single-scale segmentation methods was ameliorated. The accuracy of single-tree identification was improved from 0.67 to 0.92. This method proved to be more effective in the segmentation of forest trees using remote sensing, demonstrating high applicability across various forest types. It established a solid data foundation for subsequent single-tree information extraction and held promising prospects for practical applications.