SATELLITE IMAGE ANALYSIS AND CROP RECOMMENDATION SYSTEM

Abstract

Precision agriculture demands scalable tools to analyse vast tracts of land rapidly. AgroVision is a web-based crop recommendation framework that fuses YOLOv8 satelliteimage object detection with a rule-based Explainable AI (XAI) engine to map land types and prescribe contextually appropriate crops. Satellite imagery — including Sentinel-2 true-colour composites and high-resolution aerial photographs — is processed through a pre-trained YOLOv8 nano model (best.pt, 23 MB) which detects eight land-cover classes: water, water body, agriculture, forest, urban, dry land, river, and pond. The detected classes are passed to the XAI recommendation engine, which evaluates four factors — season suitability, soil compatibility, water requirements, and economic viability — to generate a confidence-ranked list of primary and alternative crops. The system achieves a mean [email protected] of 0.951 across all land classes and delivers subsecond inference latency via a Flask–MySQL web platform serving registered users and an administrator dashboard for prediction audit.