The model, named SRD-YOLO, achieved a mean average precision of 96.5% and an F1 score of 94%, outperforming existing methods under challenging conditions such as occlusion, dense weed clusters, and variable lighting.
Corn is one of the world’s most important crops, but its early growth stages are highly vulnerable to weed competition. Weeds deprive young corn plants of water, nutrients, and light, leading to lasting reductions in yield. Traditional detection methods can identify weeds but often miss their critical apical meristem—the growth point that enables regrowth. Without precise targeting, technologies such as laser or robotic weeding lose efficiency, resulting in wasted resources and environmental impact. Based on these challenges, there is an urgent need for accurate methods that can locate weed growth points and guide precision agriculture systems toward smarter, more sustainable weed control
A study (DOI: 10.1016/j.plaphe.2025.100072) published in Plant Phenomics on 20 June 2025 by Yinfa Yan’s team, Shandong Agricultural University, paves the way for real-time, 24-hour precision weeding robots that target weeds at their most vulnerable stage, reducing herbicide use and supporting sustainable agriculture.
To achieve this, the researchers built a comprehensive dataset of 3,380 images capturing corn seedlings and seven common weed species under varying field conditions, including strong daylight, weak light, and nighttime illumination with LEDs. They annotated growth points as keypoints and weeds as bounding boxes, ensuring that the model learned to detect precise targets. Building on the YOLOv8-Pose architecture, the team designed three key improvements. First, a dilation-wise residual module (DWRM) was added to extract multiscale contextual information, enhancing accuracy when growth points were small or obscured. Second, a RepViT block (RVB) optimized the network for lightweight deployment, reducing computational load and model parameters by 8.7 million while speeding up processing. Third, a separation and enhancement attention module (SEAM) was introduced to emphasize weed features and recover occluded details. The resulting model, SRD-YOLO, was tested against YOLOv5-Pose, YOLOv6-Pose, YOLOv7-Pose, and YOLOv8-Pose. Across multiple weed species—including Eleusine indica, green bristlegrass, and purslane—SRD-YOLO consistently outperformed its competitors. For example, it achieved F1 scores above 90% for all species, even in dense weed clusters or poor lighting. In ablation studies, each added module improved accuracy and speed: SEAM boosted occlusion detection by 2.9%, DWRM improved multiscale recognition by 3.2%, and RVB increased frames per second to 169 while lowering computational costs. Field tests on a delta-arm precision weeding robot confirmed the model’s practical potential, guiding targeted herbicide spraying during both day and night operations. Even under resource-limited edge computing conditions, the system achieved real-time detection speeds of 24 FPS at 0.3 m/s walking speed
This study demonstrates that AI-powered keypoint detection can transform weed management by focusing directly on growth points—the Achilles’ heel of weeds. The authors emphasize that SRD-YOLO not only improves detection accuracy but also enables lightweight deployment in real-world environments, addressing both performance and resource constraints. Looking ahead, the team plans to refine the system to handle nighttime motion blur and further improve classification of visually similar species. With continued advances, the method could significantly reduce herbicide use, enhance yields, and support environmentally friendly farming practices.
###
References
DOI
10.1016/j.plaphe.2025.100072
Original URL
https://doi.org/10.1016/j.plaphe.2025.100072
Funding information
This work was supported in part by the Tibet Shigatse Science and Technology Projects (No. RKZ2024ZY-03), the Shandong Province Modern Agricultural Industry Technology System, China (No. SDAIT-18-06), the China Agriculture Research System of MOF and MARA (No. CARS-18-ZJ0402), and the National Natural Science Foundation of China (No. 32001419).
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.