Related Experiment Video
Updated: May 29, 2026
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
Multi-scale spatial-temporal remote sensing fusion for phenology identification in rice germplasm resources
Huimin Wang1, Wei Guo2, Yue Mu1
1Engineering Research Center of Plant Phenotyping, Ministry of Education, Collaborative Innovation Center for Modern Crop Production co-sponsored by Province and Ministry, Academy for Advanced Interdisciplinary Studies, Nanjing Agricultural University, Nanjing, 210095, China.
This study introduces a deep learning framework for precise rice phenology monitoring using multi-resolution UAV imagery. The method enhances yield prediction and germplasm evaluation by integrating medium and high-resolution data, reducing costs and improving accuracy.
Area of Science:
- Agricultural Science
- Remote Sensing
- Deep Learning
Background:
- Precise crop phenology monitoring is crucial for yield prediction and germplasm evaluation in large-scale rice breeding.
- Unmanned Aerial Vehicle (UAV) remote sensing faces challenges in phenological monitoring due to cultivar asynchrony and resolution-temporal trade-offs.
Purpose of the Study:
- To develop a multi-scale temporal deep learning framework for accurate rice phenology monitoring.
- To address challenges of phenological asynchrony and data fusion from multi-resolution UAV imagery.
Main Methods:
- Proposed a deep learning framework integrating high-frequency medium-resolution (MR) images with sparse high-resolution (HR) images.
- Introduced a Missing Aware Gated Fusion (MAGF) mechanism for dynamic multi-resolution feature integration on non-aligned timelines.
- Validated the framework on a large dataset of approximately 500 rice cultivars over two growing seasons.
Main Results:
- The multi-scale fusion framework significantly outperformed single-temporal-scale baselines, achieving an Overall Accuracy (OA) of 0.873 and F1-score of 0.84.
- A hybrid sampling strategy (daily MR + weekly HR) reduced flight time by 79% while maintaining high accuracy.
- The model demonstrated strong generalization, maintaining high accuracy when applied to a different growing season.
Conclusions:
- The proposed framework offers a flexible, scalable, and cost-effective solution for high-throughput phenotyping in precision breeding.
- This approach enables robust phenological modeling despite irregular sampling and coexisting growth stages.
- The study highlights the potential of integrating multi-resolution remote sensing data for advanced agricultural monitoring.
