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 Cassava

Nigeria loses an estimated 40% of annual crop production to diseases, pests, and post-harvest spoilage; a burden costing the sector over USD 9 billion annually. Smallholder farmers, who contribute more than 70% of national food output, lack access to timely and affordable disease diagnosis and yield advisory services. This paper introduces PRECIA-NG (Precision Crop Intelligence and Analytics for Nigeria), a multi-stage AI pipeline that fuses multispectral unmanned aerial vehicle (UAV) imagery with deep learning to simultaneously detect crop diseases and forecast yields across four principal staple crops: maize, rice, sorghum, and cassava. PRECIA-NG deploys YOLOv8m fine-tuned on a 15,200-image Nigerian field dataset for disease localisation, EfficientNet-B4 for leaf-level classification, and a stacked ensemble (LSTM + XGBoost + Random Forest with Ridge meta-learner) for Local Government Area (LGA)-level yield forecasting. On held-out test sets, PRECIA-NG achieves a disease detection macro-F1 of 89.6% and mAP@0.5 of 93.2%, alongside yield prediction R2 values of 0.864-0.931 with RMSE of 0.28-0.41 t/ha. A federated learning layer built on FedAvg enables privacy-preserving model aggregation across Nigeria's six geopolitical zones without centralising farmer data. The framework delivers an average 14-day early warning window, positioning PRECIA-NG as a deployable advisory tool for FMARD, NAERLS extension services, and TETFund-sponsored agricultural research programmes.