A novel deep learning framework for field-scale wheat yield prediction.

Area of Science:

• Agricultural Science
• Machine Learning
• Remote Sensing

Background:

• Proximal sensing technologies provide rapid, non-destructive crop monitoring using spectral vegetation indices.
• Accurate pre-harvest wheat yield prediction is crucial for agricultural management and breeding programs.

Purpose of the Study:

• To develop and evaluate a genetic algorithm-optimized deep neural network for field-scale wheat yield prediction.
• To compare the performance of the proposed deep learning model against traditional machine learning algorithms.

Main Methods:

• Utilized proximal sensing data including Normalized Difference Vegetation Index (NDVI), Canopy Temperature (CT), and Plant Height (PH).
• Developed a Deep Neural Network (DNN) optimized via a Genetic Algorithm (GA).
• Trained and validated the model on 3,350 diverse wheat germplasm under irrigated and rainfed conditions.

Main Results:

• The GA-optimized DNN significantly outperformed Random Forest Regression (RFR), LASSO, and SVR in wheat yield prediction.
• NDVI measurements across five growth stages demonstrated strong predictive power (R² ≥ 60% irrigated, R² ≥ 50% rainfed).
• Random Forest Regression identified key features influencing grain yield prediction.

Conclusions:

• A GA-optimized DNN using proximal sensing data offers a robust and scalable solution for accurate wheat yield prediction.
• This approach represents the first application of GA-optimized DNN with proximal sensing for crop yield prediction in Indian agriculture.
• The model supports efficient genotype selection and contributes to sustainable development goals through improved yield estimation.