Evaluation and Optimization of Prediction Models for Crop Yield in Plant Factory

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Plant factory cultivation requires precise monitoring for optimal yield.
• Accurate crop canopy area estimation is crucial for yield prediction.
• Existing methods face challenges with background interference and prediction accuracy.

Purpose of the Study:

• To develop a robust method for crop canopy projection area (CCPA) measurement in plant factories.
• To identify the optimal machine learning model for accurate crop yield prediction.
• To enable efficient, real-time deployment of yield prediction models in cultivation systems.

Main Methods:

• Precise capture of crop canopy images and removal of background interference.
• Calculation of spatial resolution and accurate CCPA determination (R²=0.98).
• Comparative analysis of 28 prediction models using metrics like MSE, RMSE, MAE, MAPE, and R².

Main Results:

• The Wide Neural Network model demonstrated superior performance with R²=0.95, RMSE=27.15 g, and MAPE=11.74%.
• The optimal model achieved a prediction speed of 60,234.9 observations/second.
• The model's small size (7039 bytes) facilitates practical, real-time application.

Conclusions:

• The developed image analysis and Wide Neural Network approach significantly enhances crop yield prediction accuracy in plant factories.
• This method provides a foundation for refining cultivation processes and improving crop yields.
• The model's efficiency and accuracy support practical crop management and decision-making.