Overview

AgriGuard is a full-stack smart farming system built for agricultural leaders and farmers. When a user photographs a crop leaf, the system identifies the crop and disease using a complex two-stage deep learning pipeline, returning a complete pesticide advisory including dosage, spray interval, and safety guidelines.

Additionally, an ESP32-based custom hardware module handles real-time soil moisture monitoring and automated water pump control.

10 crops · 76 disease classes · 99.15% crop accuracy · 94.86% disease accuracy

The Problem

Crop disease identification traditionally takes days when waiting for a field expert visit, leading to potential crop loss. Furthermore, manual irrigation management requires constant physical presence and guesswork regarding soil moisture levels.

The Solution

An AI-powered mobile application integrated with custom IoT hardware. The system flags diseases in seconds with extremely high accuracy and automatically manages irrigation and alerts, eliminating guesswork and significantly reducing response times.

Key Features

Disease Detection: Two-stage EfficientNetB3 pipeline with a "crop gate" that rejects unknowns before the disease classifier runs. Images can be captured via mobile camera, gallery, or remotely via an ESP32-CAM.
Pesticide Advisory: A MongoDB-backed advisory collection covers all 76 disease classes, returning exact pesticide names, dosages, and safety precautions.
Smart Irrigation: ESP32-WROOM reads soil moisture and controls a 5V pump via a relay. Supports remote manual ON/OFF and a fully automated mode based on moisture thresholds.
Robust Authentication: OTP-based email verification, bcrypt password hashing, JWT sessions, and isolated user profiles.
Scan History: All results are persisted with crop, disease label, confidence scores, and timestamps with strict user-scoped access control.

Tech Stack & Architecture

This project is a massive monorepo comprising a mobile frontend, a REST API, an ML server, and C++ hardware code.

Machine Learning Pipeline (Flask & TensorFlow)

Two-model architecture built on EfficientNetB3 backbones.
Model 1 (Crop Classifier): Validates the image is a supported crop (99.15% accuracy).
Model 2 (Disease Classifier): Identifies the specific disease from 76 classes (94.86% accuracy).
Hosted independently on HuggingFace Spaces via a highly concurrent Flask API.

Backend Infrastructure (Node.js & MongoDB)

Built with Node.js and Express, deployed on Render.
MongoDB Atlas database for user auth, scan history, and pesticide advisories.
Acts as a secure IoT proxy for communication with local hardware devices over the network.

Mobile Application (React Native)

Built with React Native and Expo.
Manages user sessions, displays rich analysis history, and acts as a remote control for hardware pumps.

IoT Hardware (ESP32 / C++)

Custom ESP32-CAM integration for remote field photography.
Custom ESP32-WROOM integration with moisture sensors and relays. Implements dynamic IP registration at boot to ensure bulletproof connectivity without relying on local DNS.

AgriGuard: AI-Powered Crop Disease Detection