2026-10-13T08:33:01Z https://www.scidb.cn/oai

10.57760/sciencedb.37904 2026-06-01T17:46:00Z

2026-06-01 Individual tree point cloud dataset doi:10.57760/sciencedb.37904 en This dataset was collected in August 2023 within the forested area of Zhejiang A&F University using a backpack LiDAR scanning system. Multi-view scanning was employed to acquire high-density three-dimensional point cloud data of individual trees. The dataset mainly consists of subtropical evergreen broad-leaved tree species with diverse heights, crown widths, and structural characteristics.Raw point clouds were processed through background removal, individual tree segmentation, noise filtering, downsampling, and coordinate normalization. A total of 391 valid individual tree samples were obtained. Each sample contains 2,048 points with corresponding normal vectors. To support point cloud registration research, source–target point cloud pairs were generated by applying random rigid transformations, including rotations of up to 45°, translations of up to 0.5 m, and small random noise. FPFH descriptors, point correspondences, and supervision labels were further generated for registration tasks.The dataset is organized into four folders: train_raw, test_raw, train_processed, and test_processed. Data are stored in PKL format and include point coordinates, normal vectors, FPFH features, ground-truth transformation matrices, correspondence information, and training labels. Minor local sparsity caused by natural occlusion may exist in some samples; however, no missing samples or significant systematic errors were observed.This dataset is intended for forest individual-tree point cloud registration, three-dimensional structural modeling, and deep learning-based point cloud processing research. Individual tree point cloud; 3D point cloud registration; deep learning jiang xiao feng PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank