Analyzing Nitrogen Effects on Rice Panicle Development by Panicle Detection and Time-Series Tracking

Area of Science:

• Agricultural Science
• Plant Phenotyping
• Artificial Intelligence in Agriculture

Background:

• Understanding rice yield formation requires detailed observation of panicle development, particularly during the heading-flowering stage.
• Comprehensive analysis of panicle development under different nitrogen treatments is currently lacking.
• Automated phenotyping methods are needed to capture subtle developmental differences.

Purpose of the Study:

• To develop and validate an automated pipeline for detailed rice panicle phenotyping using time-series images.
• To analyze panicle development and flowering traits under varying nitrogen conditions.
• To assess the pipeline's ability to identify subtle differences in panicle development.

Main Methods:

• Proposed an automated pipeline using YOLO v5, ResNet50, and DeepSORT models for image analysis.
• Utilized time-series images for tracking individual panicle development.
• Combined automated analysis with field observation data for validation.

Main Results:

• Achieved high accuracy in panicle counting (R² = 0.96) and precise heading date estimation (0.25 days error).
• Identified significant effects of nitrogen application on rice population traits like panicle number and flowering duration.
• Revealed that higher nitrogen increased flowering duration and total duration from vigorous flowering for individual panicles.

Conclusions:

• The developed pipeline accurately quantifies rice panicle traits and their responses to nitrogen treatments.
• Automated phenotyping provides novel insights into subtle panicle development and flowering dynamics.
• Findings contribute to a better understanding of nitrogen's role in rice yield formation.