Related Experiment Video
Updated: Jul 29, 2026
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
Rapeseed Stand Count Estimation at Leaf Development Stages With UAV Imagery and Convolutional Neural Networks
Jian Zhang1,2, Biquan Zhao1,2, Chenghai Yang3
1Macro Agriculture Research Institute, College of Resource and Environment, Huazhong Agricultural University, Wuhan, China.
Frontiers in Plant Science
|June 27, 2020
Summary
Accurately estimating rapeseed stand count using leaf recognition with convolutional neural networks (CNNs) in drone imagery is feasible. The optimal timing for this method is the four- to six-leaf stage, 53 days after planting.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Plant Science
Background:
- Accurate rapeseed stand count is crucial for precision agriculture, influencing fertilization, irrigation, and yield prediction.
- Estimating plant count early in the growth cycle is challenging but vital for timely agricultural management.
- Existing methods lack automated, rapid, and accurate stand counting, especially using aerial imagery and leaf characteristics.
Purpose of the Study:
- To develop and validate a method for estimating rapeseed stand count using leaf recognition via convolutional neural networks (CNNs) in unmanned aerial vehicle (UAV) imagery.
- To determine the optimal growth stage for stand counting based on leaf development (one to seven leaves).
- To assess the impact of image patch size on the performance of leaf detection and stand count estimation.
Main Methods:
- Development of a CNN model for recognizing individual rapeseed leaves in UAV imagery.
- Estimation of rapeseed stand count based on the number of detected leaves.
- Comparison of leaf detection performance across different image sample sizes (16–48 pixels).
- Calibration of the model to account for leaf overcounting due to size variations.
Main Results:
- The CNN-based leaf count method achieved high performance (F-scores > 90%) at the four- to six-leaf stage (53 days after planting).
- An optimal image patch size of 32 pixels was identified, balancing accuracy and efficiency (relative RMSE of 2.22%).
- The method demonstrated high accuracy, estimating an average of 806 out of 812 plants correctly at the optimal stage.
Conclusions:
- It is feasible to automatically, rapidly, and accurately estimate rapeseed stand count using CNNs and UAV imagery.
- The four- to six-leaf stage, around 53 days after planting, is the optimal observation window for this technique.
- This leaf-recognition approach offers a novel perspective for phenotyping and crop management in crops with distinct early-stage leaves.
