Classification of Rice Yield Using UAV-Based Hyperspectral Imagery and Lodging Feature

Area of Science:

• Agricultural Science
• Plant Breeding
• Remote Sensing

Background:

• High-yield rice cultivation is crucial for global food security.
• Manual rice yield measurements are time-consuming, costly, and limit large-scale phenotyping.
• Efficient breeding programs require rapid and accurate yield assessment methods.

Purpose of the Study:

• To develop an accurate, large-scale, and high-throughput method for rice yield classification.
• To evaluate the potential of Unmanned Aerial Vehicle (UAV)-based hyperspectral data for rice yield estimation.
• To accelerate the rice breeding process through improved phenotyping.

Main Methods:

• Utilized hyperspectral imaging from a UAV platform to capture multi-temporal data of 13 japonica rice lines.
• Developed a rice yield classification model using the XGBoost algorithm.
• Conducted comparative experiments, including intraline and interline tests, considering lodging at the mid-mature stage.

Main Results:

• The XGBoost algorithm effectively classified rice yield using hyperspectral data.
• Lodging characteristics at the mid-mature stage significantly impacted classification accuracy.
• The developed method demonstrated potential for accurate, large-scale, non-destructive rice yield estimation.

Conclusions:

• UAV-based hyperspectral measurements combined with machine learning provide a low-cost, high-throughput phenotyping solution.
• This approach significantly improves rice breeding efficiency by accelerating yield estimation.
• The method offers a non-destructive alternative to traditional manual measurements for crop assessment.