Wheat Spike and Spikelet Detection and Counting from High-resolution Digital Imagery using YOLO with Oriented Bounding Boxes

This repository provides a portion of the dataset and associated resources for "Wheat Spike and Spikelet Detection and Counting from High-resolution Digital Imagery using YOLO with Oriented Bounding Boxes".

The dataset integrates field-acquired and publicly available imagery to capture variability in wheat canopy structure, spike morphology, and environmental conditions. Field data were collected from winter wheat breeding trials at South Dakota State University (SDSU), USA, during the 2022 and 2023 growing seasons. High-resolution RGB imagery was acquired under both top-view and side-view configurations to capture diverse perspectives of wheat canopies.

To enhance dataset diversity and model generalization, the SDSU field dataset (SD-RWHD) was combined with the Rotated Global Wheat Head Dataset (RGWHD), which spans multiple geographic regions and pedoclimatic conditions. The merged dataset contains 903 images with a total of 48,521 annotated wheat spikes, representing both dense and sparse canopy scenarios.

All wheat spikes were annotated using oriented bounding boxes (OBB), which align with spike geometry and significantly reduce background contamination compared to axis-aligned bounding boxes. Annotations were generated using a semi-automated workflow involving initial manual labeling, model-assisted pseudo-label generation, and human verification to ensure accuracy and consistency.

For spikelet-level analysis, a dedicated dataset (SD-RWSLD) was developed through a two-stage process. First, spike regions were detected and extracted from field imagery using trained YOLO-OBB models with few-pixels padding to preserve structural integrity. These extracted spike images were then annotated at the spikelet level using OBBs. Additional indoor-acquired high-resolution images were included to improve robustness under controlled conditions. The final spikelet dataset consists of 4,009 spike images with 60,404 annotated spikelets.

This dataset supports:

• Wheat spike detection in complex field environments
• Spikelet detection and counting at fine spatial resolution
• Development and benchmarking of object detection models using oriented bounding boxes
• High-throughput phenotyping workflows in precision agriculture

Data Contents

• Annotated wheat spike dataset (SD-RWHD + RGWHD integration)
• Annotated wheat spikelet dataset (SD-RWSLD)
• YOLO-OBB formatted annotation files
• Supporting metadata and documentation

Annotation Format

Annotations follow the YOLO-OBB specification, where each object is represented by:

• Class ID
• Normalized coordinates of four oriented bounding box vertices (clockwise)

Potential Applications

• High-throughput wheat phenotyping
• Crop yield estimation and trait analysis
• Computer vision model development for agricultural systems
• Precision agriculture and digital agronomy

Notes

• The repository contains both image data and corresponding annotation files. This is a single-class object detection dataset, where all instances (wheat spikes or spikelets) are labeled with class ID = 0.

URL to published work: https://www.nature.com/articles/s41598-026-55761-w#citeas

Cite this article

If you use this data, please cite as follows:

Ghimire, H., Maimaitijiang, M., Thapa, S. et al. Wheat spike and spikelet detection and counting from high-resolution digital imagery using YOLO with Oriented Bounding Boxes. Sci Rep (2026). https://doi.org/10.1038/s41598-026-55761-w