Related Experiment Video
Updated: Mar 14, 2026
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.9K
Study on automatic detection of wheat spike grain number based on deep learning.
Hecang Zang1,2, Yanjing Wang3, Shengwei Wang4
1Institute of Agricultural Information Technology, Henan Academy of Agricultural Sciences, Zhengzhou, China.
Frontiers in Plant Science
|March 13, 2026
Summary
This study introduces an automated wheat spike grain counting system using computer vision. The YOLOv8n model achieved high accuracy, outperforming other deep learning models for efficient wheat yield estimation.
Area of Science:
- Agricultural Science
- Computer Vision
- Deep Learning
Background:
- Wheat yield estimation relies on accurate spike grain counts.
- Manual counting is laborious, time-consuming, and impractical for large-scale breeding.
- Automated detection of wheat spike grains presents significant challenges.
Purpose of the Study:
- To develop a fast and automatic wheat spike grain detection system using advanced computer vision.
- To evaluate the performance of state-of-the-art deep learning models for this task.
- To provide a practical tool for wheat breeding and yield estimation.
Main Methods:
- Collected and augmented 936 wheat spike grain images to 3700 total images.
- Trained and tested six deep learning models: YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, YOLOv8x, and Faster R-CNN.
- Developed a WeChat mini program for real-time wheat spike grain counting.
Main Results:
- YOLOv8n achieved the highest precision (96.8%), recall (96.8%), mAP50 (98.9%), and mAP50-95 (58.4%).
- YOLOv8n demonstrated superior performance with fewer parameters, lower FLOPs, reduced inference time, smaller model size, and less GPU memory usage.
- YOLOv8n significantly outperformed other tested models in wheat spike grain counting accuracy and efficiency.
Conclusions:
- YOLOv8n is highly effective for automated wheat spike grain counting, meeting breeding requirements.
- The model's multi-scale feature fusion and lightweight design enhance performance.
- The developed system offers a valuable reference for automated grain detection and yield estimation in various crops.
