AI-driven phenotyping pipeline unveils how wheat spike architecture shapes grain yield

By combining advanced image segmentation models, the system achieved exceptional accuracy (mean intersection-over-union = 0.948) in measuring spike and spikelet traits, closely matching manual assessments. Using SpikePheno, the team analyzed 221 wheat cultivars from across China and discovered strong correlations between specific morphological features—such as spike area and fertile spikelet size—and key yield indicators including thousand-grain weight and spike yield.

Wheat (Triticum aestivum L.) is the world’s second most important staple crop, and improving its yield is vital for future food security. However, traditional phenotyping methods—relying on rulers and manual counting—capture only basic features and cannot reflect the complex structure of wheat spikes that influence yield, stress tolerance, and disease resistance. Existing imaging and machine learning tools have made progress but often fail to generalize across cultivars or to extract fine-scale spikelet traits. These challenges have limited the understanding of how spike morphology affects yield potential. To address this knowledge gap, researchers developed a robust and cost-effective deep learning pipeline capable of capturing subtle morphological differences with high precision.

A study (DOI: 10.1016/j.plaphe.2025.100096) published in Plant Phenomics on 14 August 2025 by Ni Jiang’s team, Chinese Academy of Sciences, not only reveals how fine-scale spike traits influence yield formation but also introduces a powerful digital tool to advance precision wheat breeding.

The study developed an AI-driven phenotyping pipeline called SpikePheno, which combines a ResNet50-UNet semantic segmentation model to isolate wheat spikes and stems from the background and a YOLOv8x-seg instance segmentation model to identify and characterize individual spikelets. Model performance was evaluated in two test sets: Test 1 (cultivars seen during training, but new images/plants) and Test 2 (entirely unseen cultivars). Across both tests, SpikePheno showed high accuracy and strong generalization, with spike segmentation reaching mean intersection-over-union values near 0.95, and spikelet detection and segmentation achieving mAP50 scores as high as 0.986. YOLOv8x-seg outperformed Mask R-CNN and PointRend in both detection and segmentation and was therefore selected for downstream analysis. To verify biological accuracy, 100 accessions grown in the 2024–2025 season were scored both manually and by SpikePheno for spike length (SL), spikelet number per spike (SNS), and fertile spikelet number (FNS); predictions and manual measurements were almost identical (r = 0.9865, 0.9753, 0.9635), confirming robustness across seasons. SpikePheno was then applied to 221 diverse wheat cultivars to extract 45 spike and spikelet traits and test how they relate to yield. Traits such as spike area (SA) and average fertile spikelet area (AFS) were strongly correlated with thousand-grain weight (TGW) and yield per spike (YPS), outperforming traditional SL. Using principal component analysis and hierarchical clustering, the team grouped spikes into six structural classes that differed significantly in TGW and YPS, and found clear temporal trends: modern cultivars favor larger, broader spikes. Regional analysis across major wheat-growing provinces in China further revealed that southern regions, such as Henan, tend to produce spikes with larger areas, larger spikelets, and higher TGW and YPS, whereas northern regions, such as Beijing and Hebei, were enriched for smaller, more compact spike types.

The SpikePheno system offers a rapid, scalable, and low-cost method for high-throughput wheat phenotyping. Its ability to identify detailed structural traits provides plant breeders with accurate parameters for selecting genotypes that combine high grain weight with optimal spike morphology. By bridging imaging data with genetic and agronomic analyses, SpikePheno lays the groundwork for precision breeding and intelligent crop improvement, enabling sustainable productivity increases in the face of growing food demand and climate challenges.

###

References

DOI

10.1016/j.plaphe.2025.100096

Original URL

https://doi.org/10.1016/j.plaphe.2025.100096

Funding information

This work was supported by the Biological Breeding-National Science and Technology Major Project (2023ZD04076) and the National Key Research and Development Program of China (2023YFF1000100).

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.