Related Experiment Video
Updated: Aug 26, 2025
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.4K
Automatic estimation of rice grain number based on a convolutional neural network.
Summary
A new deep learning system, the GN-System, automatically calculates rice grain number per panicle. This tool offers a fast, accurate, and low-cost solution for rice yield evaluation and breeding research.
Area of Science:
- Agricultural Science
- Computer Vision
- Biotechnology
Background:
- Rice grain number is crucial for yield and a key trait in breeding.
- Manual counting of rice grains is inefficient, labor-intensive, and prone to errors.
Purpose of the Study:
- To develop an automated system for calculating rice grain number per panicle.
- To leverage deep learning for accurate and efficient grain counting in rice.
Main Methods:
- Developed a whole panicle grain detection (WPGD) model using Cascade R-CNN with a feature pyramid network.
- Integrated the WPGD model into the GN-System for automated grain number calculation.
- Evaluated system performance using 124 panicle samples for stability and 12 for accuracy.
Main Results:
- The GN-System demonstrated high stability with R² = 0.810, MAPE = 8.44%, and RMSE = 16.73.
- The system achieved a mean accuracy of 90.6% in estimating grain number.
- The deep learning approach proved effective for grain recognition and location.
Conclusions:
- The GN-System provides a rapid, accurate, and cost-effective tool for rice yield estimation.
- This technology supports advancements in rice breeding, genetic research, and phenotypic analysis.
- Automated grain counting significantly improves efficiency over manual methods.
