Enhancing rapeseed biomass and yield estimation with ensemble learning and synergistic multidimensional features

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Accurate rapeseed yield and biomass estimation is critical for precision harvesting.
• Limited research exists on structured rapeseed biomass and yield estimation.
• This study addresses this gap using data from Jiangsu Province.

Purpose of the Study:

• To develop accurate and interpretable models for rapeseed biomass and yield estimation.
• To identify optimal feature combinations and machine learning techniques for this purpose.
• To establish a framework for predicting rapeseed harvest characteristics.

Main Methods:

• Utilized multispectral and RGB images from unmanned aerial vehicles (UAVs) during key growth stages.
• Extracted multidimensional features including spectral, textural, and structural data.
• Developed biomass-yield estimation models using four machine learning techniques and ensemble learning.
• Employed Shapley additive explanation (SHAP) for feature contribution analysis.

Main Results:

• Spectral-texture features were most effective for biomass estimation.
• Three-dimensional (3D) spectral-textural-structural features were optimal for yield estimation.
• Ensemble learning with these features significantly improved estimation accuracy (biomass R²=0.72, yield R²=0.68).
• The model demonstrated stable predictions across variety-density interactions.

Conclusions:

• The proposed framework provides an accurate and generalizable approach for rapeseed biomass and yield estimation.
• This method offers valuable insights for precision harvesting applications.
• The study highlights the effectiveness of integrating multidimensional features and ensemble learning.