Smart Agro AI: Integration of Deep Learning and Local Farmer Knowledge for Smart Crop Recommendation & Yield Prediction

Abstract

This study presents “Smart Agro AI”, an intelligent decision-support system for precision agriculture that integrates deep learning, satellite-based vegetation indices, and farmer knowledge fusion to enhance crop recommendation and yield prediction. The system leverages Google Earth Engine (GEE) and Copernicus Sentinel-2 satellite imagery to extract live Normalized Difference Vegetation Index (NDVI) and Soil Adjusted Vegetation Index (SAVI) values, providing real-time insights into field health and crop conditions. Static soil and weather parameters obtained from Kaggle datasets complement these dynamic satellite inputs to form a comprehensive feature set. The backend employs a Deep Neural Network (DNN) for crop recommendation and a regression-based DNN model for yield prediction, both trained using scaled and encoded datasets. A knowledge integration layer fuses AI predictions with farmer survey data through a weighted approach (70% AI, 30% farmer), ensuring that recommendations are both accurate and contextually relevant. The frontend, built with Streamlit, offers an interactive and explainable interface, enabling users to input parameters, visualize predictions, and download detailed field health reports in PDF format. This integrated framework bridges the gap between machine intelligence and local expertise, enhancing decision-making in sustainable agriculture while promoting transparency, adaptability, and user trust.