HORTIVISION-NG: Multispectral UAV Imaging and Deep Learning for Disease Detection and Yield Forecasting in Tomato, Pepper, Onion, and Watermelon in Nigeria
Domaine:
agricultureType de record:
datasetmodel
Créateur:
SulAbd
Éditeur:
Hôte:
Nigeria is among Africa's largest producers of tomatoes, peppers, onions, and watermelons, yet annual diseaseinduced yield losses exceed 35% due to late detection and inadequate field surveillance. This paper presents HORTIVISION-NG, a multispectral UAV-based intelligent monitoring system that integrates EfficientNet-B4 deep learning with five spectral vegetation indices — NDVI, GNDVI, NDRE, SAVI, and RVI — for simultaneous disease classification and yield forecasting across four high-value horticultural crops. A dataset of 31,847 annotated UAV images collected from 24 farms across Kano, Kaduna, and Jigawa States covers 12 disease classes and 4 healthy baselines. The system achieves an overall macro F1-score of 93.9% for disease detection and a yield prediction R² of 0.885. Federated learning (FedAvg) enables privacy-preserving cross-farm model aggregation across 8 pilot sites. Real-time inference runs at 23.4 FPS on NVIDIA Jetson Nano. Pilot farm trials demonstrate a 27.6% average yield improvement and $312 per-hectare net economic benefit.
Similaires
PRECIA-NG: A PRECISION CROP INTELLIGENCE AND ANALYTICS FRAMEWORK FOR NIGERIA Multispectral UAV Imaging, YOLOv8 Disease Detection, and Stacked Ensemble Yield Prediction for Maize, Rice, Sorghum, and CassavaRice Plant Disease Detection and Diagnosis using Deep Convolutional Neural Networks and Multispectral ImagingAdvances in UAV-based deep learning for cassava disease monitoring and detection: A comprehensive review of models, imaging techniques, and agricultural applicationsDeep Learning based Tomato Disease Detection and Remedy Suggestions using Mobile ApplicationUAV method based on multispectral imaging for field phenotypingHyperspectral Imaging-Based Deep Learning Framework for Early Maize Leaf Disease Detection