PomeGuard: A Multi-Agent Autonomous System for Non-Destructive Phytochemical Inference and Disease Classification in Bhagwa Pomegranate Horticulture

Abstract

The Bhagwa pomegranate’s real strength lies in its rich mix of phytochemicals—think anthocyanins, punicalagins, and ellagic acid. But fungal and bacterial diseases, along with tough environmental conditions, strip away those valuable com-pounds. Right now, figuring out fruit quality usually means running destructive and time-heavy lab tests, which really slows down decision-making for farmers.
That’s why we built PomeGuard. It’s an autonomous system that checks for diseases and measures the fruit’s nutritional value without tearing it apart. The system combines a CNN based on EfficientNetB0 for spotting disease visually, and an OWL 2 DL ontology that links disease severity and environmental data to phytochemical loss inside the fruit. Everything’s managed by a LangGraph StateGraph, which organizes four agents to pull together a complete nutritional scorecard.
We tested the vision module on more than 5,000 images and it nailed classification—98.9% accuracy across five categories: Healthy, Bacterial Blight, Anthracnose, Cercospora Fruit Spot, and Alternaria Fruit Spot. By tying together computer vision and semantic reasoning, PomeGuard brings real-time, non-destructive quality checks to the field. Farmers get smarter har-vests and better resource management—exactly what precision horticulture needs.

Keywords