Smart segmentation: New LiDAR-Based AI model transforms rubber tree monitoring

Using UAV-based LiDAR data, the model overcomes long-standing challenges in distinguishing individual trees within dense canopies, delivering precise measurements of structural traits such as height, crown diameter, and volume. By introducing a cosine feature extraction network and a dual-channel clustering mechanism, RsegNet improves segmentation accuracy, even in environments with overlapping branches and complex vegetation.

Rubber trees (Hevea brasiliensis) are vital for producing natural rubber and maintaining tropical ecosystem balance. In Hainan Province, China’s largest rubber-producing region, plantations are frequently affected by typhoons and cold spells that damage tree structure and yield. Accurate monitoring of individual trees is essential to assess damage, evaluate resistance, and improve breeding programs. Traditional ground-based measurements are slow and limited, while conventional LiDAR segmentation methods struggle to differentiate trees in high-density environments due to overlapping canopies and background interference. Based on these challenges, researchers sought to develop a UAV LiDAR and deep learning–based solution to achieve high-precision tree segmentation and structure extraction for improved plantation management.

A study (DOI: 10.1016/j.plaphe.2025.100090) published in Plant Phenomics on 16 July 2025 by Guoxiong Zhou & Xiangjun Wang’s team, Central South University of Forestry and Technology, provides robust support for precision agriculture, environmental assessment, and sustainable forest management in tropical ecosystems.

The study evaluated the proposed RsegNet rubber tree segmentation framework through a controlled experimental pipeline and a series of quantitative tests. Training was conducted on a high-performance computing platform (Intel Xeon Platinum CPU, RTX 3090 GPU, Ubuntu 18.04, Python 3.8, PyTorch 1.8.1) for 150 epochs. Model performance was assessed using point-level intersection-over-union (IoU), where predicted tree instances were matched to ground truth to derive true positives, false positives, and false negatives, enabling calculation of precision, recall, and F-score. Tree height accuracy was further evaluated using root mean square error (RMSE) between predicted and reference heights. Under these settings, RsegNet delivered the best or near-best scores across all metrics on a rubber tree test set, achieving an overall F-score of 86.1%, clearly outperforming existing methods. The CosineU-Net backbone improved boundary discrimination in overlapping canopies, raising mean IoU to 67.9% (5.8 percentage points higher than traditional Minkowski convolution), while the dual-channel clustering module, which fuses breadth-first spatial clustering and Mean Shift feature clustering, increased precision to 88.5% and addressed background interference more effectively than Region Growing or HDBSCAN. Sensitivity analysis showed that optimal segmentation was obtained with a cosine similarity threshold of 0.8, a clustering distance threshold of 0.6, and a wide adaptive search range in the dynamic clustering optimization algorithm, confirming robustness across densities and tree sizes. Comparative trials against state-of-the-art networks and validation on five independent forest regions further demonstrated generalization, with RsegNet reaching up to 94.9% F-score in structured coniferous stands and maintaining high stability when estimating structural traits such as tree height (R² = 0.98, RMSE = 0.29), crown diameter (R² = 0.86), and crown volume (R² = 0.90) in real rubber plantations.

Beyond rubber plantations, RsegNet’s framework holds potential across multiple forestry and agricultural applications. The dual-channel clustering mechanism enables accurate mapping of individual trees in mixed forests, supporting carbon stock estimation and biodiversity studies. Its adaptive structure can be extended to orchards, urban greening projects, and ecosystem restoration monitoring. For tropical agriculture, RsegNet offers a new avenue for precision management—allowing automated detection of storm damage, disease symptoms, and growth variation at the individual-tree level. The model also provides critical data for developing wind-resistant rubber tree varieties and improving climate resilience.

###

References

DOI

10.1016/j.plaphe.2025.100090

Original URL

https://doi.org/10.1016/j.plaphe.2025.100090

Funding information

This work was supported by the Hainan Province Science and Technology Special Fund (Grant No. ZDYF2025XDNY113); the Central Public-interest Scientific Institution Basal Research Fund (Grant No. 1630032022007); the Special Fund for Hainan Excellent Team "Rubber Genetics and Breeding" (Grant No. 20210203); the Hunan Provincial Natural Science Foundation Project (Grant No. 2025JJ50385); and in part by the National Natural Science Foundation of China (Grant No. 62276276).

About Plant Phenomics

Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.