MFCPopulus: A Point Cloud Completion Network Based on Multi-Feature Fusion for the 3D Reconstruction of Individual Populus Tomentosa in Planted Forests

Hao Liu, Meng Yang, Benye Xi, Xin Wang, Qingqing Huang, Cong Xu, Weiliang MengForests 2025 · DOI

TL;DR

This paper introduces MFCPopulus, a novel deep learning network for reconstructing complete 3D tree models from incomplete point cloud data, a crucial task for precision forestry. By fusing multi-scale features and employing a biologically informed approach, MFCPopulus significantly improves the accuracy and detail of tree crown reconstruction, outperforming existing methods.

What Problem Does This Paper Address?

Accurately reconstructing the 3D structure of individual trees is vital for understanding forest ecosystems, quantifying biomass, and managing forests effectively. However, obtaining complete 3D point cloud data of trees, especially in dense plantations, is challenging due to factors like canopy occlusion, LiDAR limitations, and the sheer volume of data. Existing point cloud completion methods often struggle with the significant density differences between tree trunks and canopies, leading to incomplete or inaccurate reconstructions. This paper tackles the problem of reconstructing complete and detailed 3D tree models from sparse and occluded point cloud data, specifically focusing on Populus Tomentosa in planted forests.

Methods

The core of the proposed solution is the MFCPopulus network, which leverages a Multi-feature Fusion Completion approach. Key innovations include:

  • Hierarchical Adversarial Framework: This framework integrates multi-scale feature extraction using Farthest Point Sampling (FPS) at varying rates. This helps to balance the density disparities between the trunk and canopy.
  • Biologically Informed Normalization: This technique aims to preserve species-specific morphological patterns while optimizing computational efficiency.
  • Structural Characteristics Split-Collocation (SCS-SCC): This strategy prioritizes crown reconstruction by adaptively adjusting sampling ratios, leading to improved canopy coverage.
  • Cross-Layer Feature Integration: This allows for the simultaneous recovery of both global tree contours and fine-grained branch topology.
  • Generative Adversarial Network (GAN) Architecture: The network uses a generator to produce point clouds and a discriminator to ensure the realism and accuracy of the generated structures, guided by a comprehensive loss function that includes Chamfer distance and adversarial loss.

Datasets

  • Training and Testing Dataset: A dataset of 1050 manually segmented individual Populus Tomentosa trees from a plantation was used. This dataset was curated with expert validation for trunk-canopy separation and augmented with simulated occlusion scenarios.
  • Generalization Testing Dataset: The FOR-species20K dataset was used to evaluate the model's performance on other tree species, including broadleaf (Fagus sylvatica, Acer campestre) and coniferous (Pinus sylvestris) species.

Key Insights and Ideas

  • Density Disparity is Key: The significant difference in point cloud density between tree trunks and canopies is a major hurdle for existing methods. MFCPopulus's multi-scale sampling and fusion strategy directly addresses this.
  • Biological Priors Matter: Incorporating biologically informed normalization helps the model learn and preserve the natural shapes and structures of trees, leading to more realistic reconstructions.
  • Crown Reconstruction is Paramount: The SCS-SCC strategy highlights the importance of prioritizing crown reconstruction, as this is often the most occluded and complex part of the tree.
  • Generalizability is Achievable: The framework's ability to generalize to different species, even those with different foliage densities and growth patterns, is a significant step towards a universal tree reconstruction solution.

Key Results

  • Reduced Discrepancies: MFCPopulus reduced Chamfer distance variance by 23% and structural complexity discrepancies (ΔD♭) by 33% compared to state-of-the-art methods.
  • High Canopy Coverage: Achieved 94.5% canopy coverage in reconstructed outputs.
  • Spatial Alignment: Octree analysis showed 89-94% spatial alignment with ground truth across various height ratios (HR = 1.25-5.0).
  • Visual Fidelity: Visual comparisons showed MFCPopulus results closely aligning with ground truth, preserving fine details and overall tree structure.
  • Efficiency: The model demonstrated superior efficiency in utilizing input point cloud information, showing a nearly twofold increase in effective point cloud cell count compared to the input.
  • Robustness to Data Scarcity: The model maintained performance with a controlled increase in Chamfer distance (14.3%) even when trained on only 30% of the original dataset.

Limitations and Future Work

  • Uniform Spacing Assumption: The current MFCPopulus framework was primarily developed for Populus Tomentosa plantations with uniform spacing.
  • Future Work Suggestions:
    • Investigate performance in mixed-species or irregularly spaced forests.
  • Adapt the model for conifers with denser foliage by adjusting canopy sampling rates.
  • Integrate spacing metrics (e.g., neighborhood density index) and species-specific priors to enhance generalizability.
  • Further explore the impact of competitive interactions in multi-species stands on trunk bifurcation patterns.

Summary updated

Abstract

The accurate point cloud completion of individual tree crowns is critical for quantifying crown complexity and advancing precision forestry, yet it remains challenging in dense plantations due to canopy occlusion and LiDAR limitations. In this study, we extended the scope of conventional point cloud completion techniques to artificial planted forests by introducing a novel approach called Multi−feature Fusion Completion of Populus (MFCPopulus). Specifically designed for Populus Tomentosa plantations with uniform spacing, this method utilized a dataset of 1050 manually segmented trees with expert−validated trunk−canopy separation. Key innovations include the following: (1) a hierarchical adversarial framework that integrates multi−scale feature extraction (via Farthest Point Sampling at varying rates) and biologically informed normalization to address trunk−canopy density disparities; (2) a structural characteristics split−collocation (SCS−SCC) strategy that prioritizes crown reconstruction through adaptive sampling ratios, achieving a 94.5% canopy coverage in outputs; (3) a cross−layer feature integration enabling the simultaneous recovery of global contours and a fine−grained branch topology. Compared to state−of−the−art methods, MFCPopulus reduced the Chamfer distance variance by 23% and structural complexity discrepancies (ΔDb) by 33% (mean, 0.12), while preserving species−specific morphological patterns. Octree analysis demonstrated an 89−94% spatial alignment with ground truth across height ratios (HR = 1.25−5.0). Although initially developed for artificial planted forests, the framework generalizes well to diverse species, accurately reconstructing 3D crown structures for both broadleaf (Fagus sylvatica, Acer campestre) and coniferous species (Pinus sylvestris) across public datasets, providing a precise and generalizable solution for cross−species trees’ phenotypic studies.
BibTeX
@inproceedings{liu2025mfcpopulus,
  author = {Hao Liu and Meng Yang and Benye Xi and Xin Wang and Qingqing Huang and Cong Xu and Weiliang Meng},
  title = {{MFCPopulus: A Point Cloud Completion Network Based on Multi-Feature Fusion for the 3D Reconstruction of Individual Populus Tomentosa in Planted Forests}},
  booktitle = {Forests},
  year = {2025},
  doi = {10.3390/f16040635},
}
Figure