Self-adaptive individual tree modeling based on skeleton graph optimization and fractal self-similarity

Area of Science:

• Forestry and Ecological Modeling
• Computer Vision and Graphics
• Geospatial Analysis

Background:

• Accurate 3D individual tree modeling is vital for forest ecology.
• Existing methods struggle with canopy branch connectivity and data gaps.
• Challenges include inaccurate models due to incomplete point cloud data.

Purpose of the Study:

• To develop an innovative 3D individual tree modeling method.
• To address challenges of wrongly connected branches and data gaps in tree modeling.
• To improve the fidelity and robustness of tree models.

Main Methods:

• Skeleton point extraction via Laplacian-based contraction and farthest distance spherical sampling.
• Skeleton point adjustment and optimization for centralization, especially with incomplete data.
• Novel edge weight definition for correct skeleton line construction and fractal self-similarity for model refinement.

Main Results:

• The proposed method achieved tree volumes closest to reference values.
• Demonstrated a relative mean deviation of 0.01% and relative root mean square error of 0.09%.
• Achieved a concordance correlation coefficient of 0.994, outperforming TreeQSM and AdQSM.

Conclusions:

• The novel method significantly improves 3D individual tree modeling accuracy and robustness.
• It effectively handles data gaps and complex canopy structures.
• This advancement is crucial for precise forest ecological applications.