Exploring Soybean Flower and Pod Variation Patterns During Reproductive Period Based on Fusion Deep Learning

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Soybean flower and pod drop rates significantly impact crop yield.
• Accurate, high-throughput phenotyping of soybean reproductive structures is crucial for yield studies.
• Traditional manual counting methods are labor-intensive and time-consuming.

Purpose of the Study:

• To compare deep learning algorithms for identifying and counting soybean flowers and pods.
• To develop and optimize an improved Faster R-CNN model for soybean phenotyping.
• To propose a fusion model for robust soybean flower and pod recognition and counting.

Main Methods:

• Evaluation of various deep learning algorithms for object detection.
• Optimization of the Faster R-CNN model tailored for soybean flower and pod characteristics.
• Development of a fusion model integrating optimized Faster R-CNN for enhanced counting accuracy.

Main Results:

• The Faster R-CNN model demonstrated superior performance in identifying soybean flowers and pods.
• The optimized model achieved high accuracy: 94.36% for flowers and 91% for pods.
• The fusion model showed excellent agreement with manual counts (R² = 0.965 for flowers, R² = 0.98 for pods).

Conclusions:

• The developed fusion model is a robust and efficient algorithm for soybean flower and pod counting.
• This automated approach significantly reduces labor intensity and improves research efficiency.
• The model facilitates in-depth studies of soybean reproductive variability and drop patterns.