Related Experiment Video
Updated: Jul 3, 2026
07:27
Live Confocal Imaging of Developing Arabidopsis Flowers
Published on: April 1, 2017
FQGR-net: Morphology-based litchi flower quantification and gender recognition.
Jun Li1,2,3, Zhe Ma1, Jiaquan Lin1
1College of Engineering, South China Agricultural University, Guangzhou, 510642, China.
Plant Phenomics (Washington, D.C.)
|July 2, 2026
Summary
This study introduces the Flower Quantification and Gender Recognition Network (FQGR-Net) to accurately count male and female litchi flowers, addressing biennial bearing issues in agriculture. The developed system achieves high accuracy, enabling better crop management and yield optimization.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Science
Background:
- Litchi (Litchi chinensis) cultivation in China is significantly impacted by biennial bearing, leading to alternating high- and low-yield cycles.
- Unstable floral initiation is identified as the primary cause of irregular fruiting in mid-to-late maturing litchi cultivars.
- Accurate quantification of female-to-male flower ratios is crucial for implementing targeted management strategies to improve fruit set.
Purpose of the Study:
- To develop an automated system for accurate and rapid quantification of female and male flowers in litchi.
- To propose a novel deep learning model, the Flower Quantification and Gender Recognition Network (FQGR-Net), for simultaneous flower classification and counting.
- To enhance litchi crop management through precise floral phenotyping.
Main Methods:
- A three-branch neural network architecture, FQGR-Net, was designed for simultaneous classification and counting of female and male litchi flowers.
- Module-level optimization was performed to enhance both counting accuracy and computational efficiency of the FQGR-Net model.
- A self-constructed dataset was utilized for training and evaluating the model, alongside comparative experiments on public datasets.
Main Results:
- FQGR-Net achieved an average Mean Absolute Error (MAE) of 8.498 and Root Mean Square Error (RMSE) of 13.209 across categories on the self-constructed dataset.
- Regression analysis showed high correlation with ground truth, yielding R-squared values of 0.930 for female flowers and 0.971 for male flowers.
- Field trials demonstrated over 80% accuracy in female/male flower counting, validating the practical applicability of the developed system.
Conclusions:
- The proposed FQGR-Net model offers a significant advancement in automated litchi flower phenotyping, improving counting accuracy and computational efficiency.
- The developed litchi flower phenotyping analyzer addresses a technological gap, providing a tool for precise floral census and crop management.
- Accurate floral quantification using FQGR-Net can lead to optimized floral development and enhanced fruit-setting rates, benefiting litchi production.
