sAI-Powered TM-WSNet boosts precision rubber tree monitoring through 3D point cloud segmentation

By fusing Transformer and Mamba architectures with wavelet-based spatial feature extraction, the model excels at distinguishing individual trees from overlapping canopies and background vegetation. The system’s adaptive scale optimization further enhances learning efficiency across different spatial resolutions. When tested on real-world plantation datasets, TM-WSNet precisely estimated key growth parameters—including tree height, crown width, and trunk diameter—with coefficients of determination (R²) up to 1.00, 0.99, and 0.89.

Rubber trees are a critical industrial crop that supplies natural rubber to countless sectors, from automotive manufacturing to healthcare. Monitoring their growth and health requires accurate assessment of parameters such as crown diameter, trunk thickness, and canopy density. Yet, due to overlapping canopies, irregular shapes, and background noise in plantation environments, existing methods often fail to achieve reliable single-tree segmentation. Traditional machine learning approaches and early deep learning models have struggled with feature loss and boundary confusion. To overcome these long-standing challenges, researchers have turned to 3D point cloud data combined with new architectures capable of capturing both global and local spatial features.

A study (DOI: 10.1016/j.plaphe.2025.100093) published in Plant Phenomics on 21 August 2025 by Guoxiong Zhou & Xiangjun Wang’s team, Central South University of Forestry and Technology, not only advances precision monitoring in the rubber industry but also lays the foundation for intelligent yield prediction and large-scale plantation management.

The study evaluated TM-WSNet through a structured experimental pipeline that included custom and public datasets, controlled training environments, defined performance metrics, and extensive benchmarking. First, the model was trained and tested on a self-built RubberTree dataset of 990 individually labeled rubber trees from Hainan, China, and further validated on standard point cloud benchmarks (ShapeNetPart, GBS, and ForestSemantic) that span objects, single trees, and full forest scenes. Data preprocessing involved isolating individual tree targets, suppressing ground/background noise, and augmenting samples through rotation, scaling, and flipping to produce robust training sets. The model was implemented in PyTorch on high-performance GPU hardware and tuned using 200 training epochs, batch size 4, and an experimentally optimized multiscale grid configuration (“grid_3”), which best balanced fine local detail and global canopy structure. Performance was assessed using standard metrics for segmentation quality—mIoU, mAcc, mF1, precision, recall—and R² for structural parameter fitting. Under 5-fold cross-validation, TM-WSNet achieved an average mIoU of 90.35%, showing higher accuracy and stability than the backbone model without its key modules. Module-level tests confirmed that SGTramba improved boundary recognition under canopy overlap, WGMS enhanced segmentation of irregular tree forms, and the SCPO learning-rate optimizer boosted convergence across feature scales. Ablation experiments showed that integrating these modules raised mIoU by up to 9.59% and improved mAcc, mF1, and mPrec. Against state-of-the-art baselines such as PointTransformerV2, PointMamba, and PC-Mamba, TM-WSNet achieved the best scores on RubberTree (mIoU 88.78%, mAcc 94.50%, R² ≈ 0.985) and remained competitive on ForestSemantic, ShapeNetPart, and GBS. Finally, in real UAV LiDAR scans of plantations, the model maintained high performance (mIoU 86.15%, mAcc 92.56%), accurately estimating tree height, crown width, and trunk diameter, demonstrating both field readiness and generalization.

The TM-WSNet system provides a powerful tool for precision agriculture and digital forestry, offering accurate tree-level data for yield forecasting, carbon stock estimation, and health diagnosis. Its high automation and strong generalization capability make it suitable for large-scale monitoring using LiDAR-equipped UAVs or autonomous vehicles. By enabling efficient extraction of key parameters such as canopy volume and trunk diameter, TM-WSNet supports evidence-based plantation management, sustainable resource utilization, and improved productivity in the global rubber industry.

###

References

DOI

10.1016/j.plaphe.2025.100093

Original URL

https://doi.org/10.1016/j.plaphe.2025.100093

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.