ForestSemantic: A Dataset for Semantic Learning of Forest from Close-Range Sensing

ForestSemantic is a dataset for forest semantic studies at both tree- and plot-levels. The dataset supports both instance and semantic segmentation, such as the tree detection and segmentation and the classification of ground, trunk, branches, and foliage components at both tree- and plot-levels. Also, the instance of each first-order branch is provided,

For each plot, three files are provided, i.e., "Plot_x.las", "Plot_x_Tree_Reference.xlsx" and "Plot_x_Branch_Reference.txt", where x means the x-th plot.
1) "Plot_x.las" is the data file, which includes the point coordinates and intensity, as well as tree-, classification-, and First-order branch IDs. The tree-, classification-, and First-order branch IDs are stored in the field of "Point Source ID", "Classification" and "GPS Time", respectively.

2) "Plot_x_Tree_Reference.xlsx" includes the reference of the tree structure traits for each tree in the plot. The reference of each tree takes up one row. The tree-ID, position_x, position_y, tree height (m), DBH (m), First-order branch (m), Crown Projection area (m²), Crown Surface area (m²), Crown Volume (m³) are in the column 1 to 9, respectively.

3) "Plot_x_Branch_Reference.txt" includes the reference of the First-order branch in the plot, including the tree ID, branch ID, the start and end point positions of each branch. The record of each individual First-order branch takes up one row, and the column 1 to 9 are tree-ID, First-order Branch ID, Start_x, Start_y, Start_z, End_x, End_y, End_z, and Length.

4) The calculation of the reference of the tree structure traits can be found in https://doi.org/10.1080/10095020.2024.2313325.

5) For more details about the data, readers are referred to "Read me.pdf".

If you used this dataset, please cite the following paper:

Liang, Xinlian, Hanwen Qi, Xuejie Deng, Jianchang Chen, Shangshu Cai, Qingjun Zhang, Yunsheng Wang, Antero Kukko, and Juha Hyyppä. 2024. “ForestSemantic: A Dataset for Semantic Learning of Forest from Close-Range Sensing.” Geo-Spatial Information Science, March, 1–27. doi:10.1080/10095020.2024.2313325.