By integrating advanced deep learning with efficient clustering, the approach outperformed current techniques and promises to accelerate plant phenotyping research worldwide.
Plant phenotypes—observable traits shaped by genetics and environment—are the foundation of breeding and crop science. Traditional trait measurement relies heavily on manual work, often destructive, time-consuming, and prone to error. In the past two decades, 2D imaging and later 3D sensors such as LiDAR have transformed the field, enabling digital reconstruction of plants. However, accurately distinguishing individual plant organs, especially in complex species, remains challenging. Current models often lack flexibility and struggle to generalize across different crops and growth stages. Based on these limitations, researchers sought a robust and generalized method for 3D organ segmentation that could handle structural variation in multiple species.
A study (DOI: 10.1016/j.plaphe.2025.100065) published in Plant Phenomics on 7 June 2025 by Xiuhua Li’s team, Guangxi University, opens new possibilities for high-throughput phenotyping in breeding and precision agriculture.
The researchers constructed a stem–leaf semantic segmentation model using the PointNeXt deep learning framework implemented in PyTorch 1.11 on an Ubuntu 18.04 environment, supported by an Intel i9-10900X CPU, 120 GB of memory, and an NVIDIA RTX3090 GPU. The dataset was labeled with two classes—stems and leaves—and training employed cross-entropy loss with label smoothing and the AdamW optimizer, with an initial learning rate of 0.001 and cosine decay. To optimize performance, the team tested different hyperparameters. Among multilayer perceptron (MLP) channel sizes, 64 channels provided the best balance between accuracy and efficiency, achieving higher overall accuracy and mean intersection over union (mIoU) than 32 or 160 channels. Similarly, experiments with different configurations of InvResMLP blocks behind the four SA modules showed that the setup B=(1,1,2,1) delivered the best tradeoff, with an overall accuracy of 97.03% and F1 score of 93.98%. Applying this optimized configuration, PointNeXt was evaluated on 35 sugarcane, 14 maize, and 22 tomato plants at different growth stages. Results demonstrated high accuracy across all crops, with mIoU values of 89.21%, 89.19%, and 83.05% for sugarcane, maize, and tomato, respectively, and mean overall accuracies above 94%. Sugarcane performed slightly better due to a larger training set, while tomato proved more challenging because of its dense and irregular leaf structure. The Quickshift++ clustering algorithm was then employed for leaf instance segmentation, successfully identifying leaf edges and boundaries in monocots and distinguishing individual leaflets in tomatoes. Quantitative scores exceeded 90% precision and recall for sugarcane and maize, though tomato lagged due to overlapping leaflets. Finally, comparative tests against four state-of-the-art networks (ASIS, JSNet, DFSP, and PSegNet) confirmed that the proposed two-stage method consistently outperformed existing models, achieving average values of 93.32% precision, 85.60% recall, 87.94% F1, and 81.46% mIoU across all crops.
With accurate and automated analysis of plant structures, researchers can better link genotype to phenotype, accelerate trait discovery, and optimize crop management. Beyond breeding, the method could assist in monitoring crop health, modeling growth under climate stress, and informing smart agricultural systems that rely on detailed plant data. Because it avoids destructive sampling, the approach supports sustainable and large-scale research without harming valuable specimens.
###
References
DOI
10.1016/j.plaphe.2025.100065
Original URL
https://doi.org/10.1016/j.plaphe.2025.100065
Funding information
This work was supported by the Science Technology Major Project of Guangxi, China (Gui Ke Nong AB24153010, Gui Ke AA22117004), the National Natural Science Foundation of China (31760342) and the Innovation Project of Guangxi Graduate Education, YCSW2023028.
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.