A Multimodal UAV Dataset for Rice Yield Prediction in the Vietnam Mekong Delta
Description
This repository contains a multimodal UAV dataset for rice yield prediction in the Vietnam Mekong Delta (VMD), including 184 plot-level RGB UAV images acquired at the flowering stage, grain-yield measurements, vegetation-index features, agronomic variables, EffNet-derived image features, and source code associated with the study. Vegetation-index features include GNDVI, NDRE, LCI, SIPI2, CIre, and NDYI, extracted across six phenological stages: jointing (JT), booting (BT), heading (HD), flowering (FL), milk (ML), and dough (DG). Deep spatial features were generated using the EffNet (EfficientNet-B0) architecture through a five-fold out-of-fold (OOF) framework. The repository includes complete and selected vegetation-index datasets, multimodal datasets, FL-stage datasets, EffNet feature representations, plot-level UAV images, and source code for image preprocessing, feature extraction, machine-learning models, ensemble model, and SHAP analysis. The repository is provided to support reproducible research on UAV-based rice yield prediction, multimodal machine learning, explainable artificial intelligence, and precision agriculture.
Files
Steps to reproduce
1. Download and extract the dataset repository. 2. Use the files in the 'data' directory to access vegetation-index, agronomic, multimodal, and yield data. 3. Use the 'images_raw' directory to access the 184 plot-level RGB UAV images used for EffNet feature extraction. 4. EffNet-derived image representations generated using a five-fold OOF strategy are provided in 'effnet_feature_run1.csv' to 'effnet_feature_run5.csv'. 5. The source code required to reproduce feature extraction, machine-learning models, ensemble model, and SHAP analysis is available in the 'code' directory.
Institutions
- An Giang UniversityAn Giang Province, Long Xuyen
- Vietnam National University Ho Chi Minh CityHo Chi Minh, Ho Chi Minh City