PomeGuard 2.0: A Production-Ready, Full-Stack Agricultural Intelligence System for Non-Destructive Pomegranate Disease Classification, Phytochemical Inference, and Precision Farm Management

Abstract

We present PomeGuard 2.0, the completed and production-deployed second version of our multi-agent agricultural intelligence platform for the Bhagwa pomegranate cultivar. Building upon the foundational Version 1 architecture published on The International Innovations & Scholarly Trends Journal (IISTJ) [51]—which introduced EfficientNetB0-based disease classification achieving 98.9% accuracy and OWL 2 DL ontological reasoning for phytochemical inference—Version 2 delivers a comprehensive, full-stack system suitable for real-world precision horticulture deployment. New engineering contributions include: (i) an AI-powered conversational assistant (Google Gemini) contextualised per scan result, with persistent chat memory via Firebase Realtime Database; (ii) a Farm Tools module providing disease-specific, area-scaled pesticide and fertiliser dosage calculators; (iii) a live, GPS-anchored disease hotspot map (Leaflet.js with ESRI satellite imagery) integrated into the farmer dashboard; (iv) real-time weather ingestion from the Open-Meteo API to automatically populate environmental metadata; (v) a WhatsApp Business API integration for alert delivery; (vi) a Gemini-powered image validation guard that rejects non-pomegranate uploads; and (vii) a cloud-native persistence stack—Supabase (PostgreSQL + Auth), Cloudinary, and Firebase—providing per-user data isolation, secure authentication, and permanent image archival. The vision backbone retains its 98.9% overall classification accuracy across five disease classes. The complete system is deployed via Vercel with sub-5-second end-to-end analysis latency, marking the system as production-complete and ready for field deployment.

Keywords