Estimation of soybean yield based on high-throughput phenotyping and machine learning.

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Soybean self-sufficiency in China is a significant concern, with high import volumes.
• Accurate soybean yield prediction is crucial for agricultural planning and breeding.
• RGB cameras and machine learning offer promising avenues for crop yield estimation.

Purpose of the Study:

• To develop and evaluate machine learning models for soybean yield prediction using RGB image features.
• To identify optimal features and model parameters for accurate yield estimation.
• To assess the impact of multi-angle image perspectives on prediction accuracy.

Main Methods:

• Captured RGB images of 240 soybean varieties from side and top perspectives at the R6 stage.
• Extracted morphological, color, and textural features from the images.
• Applied feature selection using Pearson correlation coefficient (≥0.5).
• Developed yield estimation models using five machine learning algorithms (CatBoost, LightGBM, RF, GBDT, MLP).

Main Results:

• Gradient Boosting Decision Tree (GBDT) emerged as the optimal model, achieving an R2 of 0.82.
• The model demonstrated strong predictive performance with an RMSE of 1.99 g/plant and MAE of 3.12%.
• Fusing multi-angle and multi-type image indicators enhanced soybean yield prediction accuracy.

Conclusions:

• Machine learning models utilizing RGB image parameters show high potential for soybean yield estimation.
• This approach provides a theoretical basis and technical support for accelerating soybean breeding.
• The integration of multi-angle imaging and advanced algorithms is key to improving agricultural predictions.