Automatic wheat ear counting using machine learning based on RGB UAV imagery.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Breeding

Background:

• Wheat ear count per area is crucial for yield determination.
• Current manual methods for wheat ear counting are time-consuming and lack standardization.
• Advancements in phenotyping and monitoring require efficient, automated evaluation techniques.

Purpose of the Study:

• To develop and validate an automated wheat ear-counting system using RGB images from unmanned aerial vehicles (UAVs).
• To compare the accuracy and efficiency of the automated system against manual counting methods.
• To assess the correlation of automated and manual ear counts with grain yield under varying nitrogen treatments.

Main Methods:

• Utilized machine learning techniques on RGB images captured by a UAV.
• Implemented a frequency filter, segmentation, and feature extraction for ear discrimination.
• Evaluated performance across 12 winter wheat cultivars with three nitrogen treatments.
• Conducted in-situ manual ear counting for secondary validation.

Main Results:

• The automated system demonstrated high accuracy and efficiency in counting wheat ears compared to image-based manual counts.
• Correlations between automated ear counts and grain yield were stronger than manual counts, particularly under low nitrogen conditions.
• The study provides insights into the methodological requirements and limitations of automated ear counting.

Conclusions:

• Automated wheat ear counting using UAV-based RGB imagery is a viable and efficient alternative to manual methods.
• This technology can significantly advance high-throughput phenotyping in wheat breeding.
• The system's strong correlation with grain yield highlights its potential for optimizing crop management and breeding strategies.