Yield and Quality Prediction of Crops using a Deep Learning-Based Multimodal Data Integration Framework

A precise prediction of crop yield and quality is essential for improving resource efficiency, optimising supply chain management, and enhancing food security, particularly in the face of climate change and resource constraints. This paper proposes a deep learning-based multimodal integration framework that combines Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and an attention mechanism to predict yield and quality jointly. The framework integrates heterogeneous data sources, including soil properties, irrigation records, microclimate variables, and vegetation indices (NDVI/EVI) derived from remote sensing imagery. Simulation experiments were conducted on rice and tomato datasets under both normal and stressful conditions, including six representative scenarios: drought stress, excessive irrigation, heat waves, fertiliser mismanagement, pest and disease outbreaks, and soil salinity. Results indicate that the proposed model consistently outperforms baseline approaches, achieving an RMSE of 245 kg/ha and an R2 of 0.93 for rice yield, and an MAE of 0.15% and an R2 of 0.92 for tomato quality. The attention mechanism further enhances interpretability by identifying critical growth stages and influential features. These findings confirm the robustness and practical relevance of the proposed framework for precision and climate-smart agriculture, providing a comprehensive tool to optimise productivity and post-harvest quality simultaneously.