Exploring UAV-Based Multispectral Crop Yield Analysis: Opportunities and Challenges

Combining UAVs with multispectral imaging technology has enabled high-resolution, real-time monitoring of crops and prediction of crop yields, revolutionizing precision farming. This research focuses on UAV yield estimation, along with machine and deep learning models with Explainable Artificial Intelligence (XAI) for multispectral imaging. The research has deployed UAVs to gather other relevant data with multispectral imaging and generated orthomosaics along with vegetation indices, which are NDVI, EVI, and GNDVI, as well as different agronomic and environmental data. Afterward XGBoost, LASSO regression predictive models, and dimensionality reduction were compared to Convolutional Neural Networks (CNNs). The spatial and spectral Ubisoft models were also incorporated, where CNNs provided the most predictive power with a validation R2of 0.8486. The XAI tools proposed, SHAP and Grad-CAM, determined which vegetation indices and spatial patterns apportioned at the field level influenced yield forecasting. The research has shown predictive AI for actionable insights in precision farming. The research aims to develop hybrid deep learning models for rapid, scalable, real-time use on a multi-crop dataset. The research proposes focusing on its real-time applications.