🌿 Krishinoor — AI Plant Disease Detection Android App

Krishinoor is an offline AI-powered Android application for real-time plant disease detection.
Built using Kotlin, PyTorch Mobile, and Google Location Services — designed for smart farming, precision agriculture, and UAV-assisted crop monitoring.

---

📱 App Structure

Login & Language	Dashboard	Scan Plant
Multilingual login with Hindi, English, Bengali, Urdu support	Real-time scan stats with GPS tracking	Camera + Gallery input with UAV drone support

Healthy Detection	Disease Detection	Scan History
Grape — Healthy (80.1%)	Pepper — Bacterial Spot (100%)	GPS-tagged scan history with confidence

---

✨ Key Features

Feature	Description
🤖 Offline AI Inference	MobileNetV2 model runs fully on-device — no internet required
🌍 Multilingual UI	Supports English, Hindi (हिन्दी), Bengali (বাংলা), Urdu (اردو)
📍 GPS Tagging	Every scan is tagged with precise GPS coordinates and grid reference
📡 UAV/Drone Support	Reads GPS from drone photo EXIF metadata automatically
📊 Scan Dashboard	Real-time stats — total scans, diseased vs healthy plant count
🕐 Scan History	Complete history with plant name, disease, confidence, location, timestamp
🔔 Push Notifications	Instant alert when disease is detected
💊 Treatment Database	Offline remedy + severity info for all 38 disease classes
📷 Camera + Gallery	Supports both live camera capture and gallery image upload
🔒 Secure Login	Phone number + password authentication with register flow

---

🎯 App Flow

Splash Screen
    ↓
Login / Register (Phone + Password)
    ↓
Dashboard (Scan stats + Last scan preview)
    ↓
Scan Plant Screen (Camera / Gallery)
    ↓
AI Inference (PyTorch Mobile — offline)
    ↓
Analysis Result
  ├── Plant Name
  ├── Disease / Healthy status
  ├── Confidence score
  ├── Severity (High / Medium / Low / None)
  ├── Treatment & Solution
  ├── GPS Location + Grid coordinate
  └── Timestamp
    ↓
Scan History (all previous scans)

---

🧠 AI Model

The app uses a fine-tuned MobileNetV2 model exported as TorchScript Lite (.ptl) for on-device inference.

Property	Value
Architecture	MobileNetV2
Classes	38 disease categories
Plants Supported	14 species
Model Size	~9.3 MB
Inference Time	~200ms on mid-range Android
Format	TorchScript Lite (.ptl)
Internet Required	❌ Fully Offline

🔗 Full ML pipeline: krishinoor-ml-model

Image Preprocessing Pipeline (Android)

Bitmap → ARGB_8888 conversion
→ Resize (smaller side = 256px)
→ Center Crop (224 × 224)
→ Normalize (ImageNet mean/std)
→ TorchScript Lite Inference
→ Argmax → Class Label

---

🛠️ Tech Stack

Layer	Technology
Language	Kotlin
ML Runtime	PyTorch Mobile (LiteModuleLoader)
Location	Google Play Services (FusedLocationProvider)
Camera	AndroidX Camera2 + FileProvider
UI	Material Design 3, ViewBinding
Storage	SharedPreferences (scan history)
Notifications	NotificationManager (Android 13+)
Image EXIF	AndroidX ExifInterface
Build System	Gradle (Kotlin DSL)
Min SDK	Android 8.0 (API 26)
Target SDK	Android 14 (API 34)

---

📋 Permissions Required

Permission	Purpose
CAMERA	Capture leaf images for analysis
ACCESS_FINE_LOCATION	GPS tagging of scan location
ACCESS_COARSE_LOCATION	Fallback location
INTERNET	Optional future cloud sync
POST_NOTIFICATIONS	Disease detection alerts

---

🗂️ Project Structure

app/src/main/
├── java/com/plantdoctor/krishinoor/
│   ├── SplashActivity.kt          # Launch screen
│   ├── LoginActivity.kt           # Phone + password login
│   ├── RegisterActivity.kt        # New user registration
│   ├── MainActivity.kt            # Dashboard with scan stats
│   ├── ScanActivity.kt            # Camera/gallery + AI inference
│   ├── ResultActivity.kt          # Analysis result display
│   ├── HistoryActivity.kt         # Scan history list
│   ├── PlantDiseaseClassifier.kt  # PyTorch Mobile inference engine
│   └── DiseaseDatabase.kt         # Offline remedy + severity data
├── assets/
│   ├── model_mobile.ptl           # TorchScript Lite model (~9.3MB)
│   └── classes.json               # 38-class label mapping
└── res/
    ├── layout/                    # XML UI layouts
    ├── values/                    # Strings (multilingual)
    └── xml/                       # FileProvider paths

---

🚀 Build & Run

Prerequisites

• Android Studio Hedgehog or later
• Android device / emulator (API 26+)
• JDK 11+

Steps

# Clone the repository
git clone https://github.com/Sneha8271/krishinoor.git
cd krishinoor

# Open in Android Studio
# File → Open → select krishinoor folder

# Build and run
# Click ▶️ Run or press Shift+F10

The model (model_mobile.ptl) and class labels (classes.json) are bundled in app/src/main/assets/ — no additional setup needed.

---

📊 Sample Inference Results

Plant	Result	Confidence
🍇 Grape	Black Rot — Diseased	100.0%
🍇 Grape	Esca (Black Measles)	97.6%
🍇 Grape	Healthy	99.4%
🫑 Pepper	Bacterial Spot	100.0%
🍅 Tomato	Early Blight	85.0%
🍎 Apple	Cedar Apple Rust	78.4%

---

🌾 Real-World Use Cases

• 🌱 Early disease detection before visible crop damage
• 📵 Offline diagnosis for farmers in low-connectivity rural areas
• 🚜 Precision agriculture — targeted treatment, reduced pesticide use
• 🛸 UAV/drone image analysis via automatic EXIF GPS extraction
• 📱 Accessible smartphone-based screening for small-scale farmers
• 🌍 Multilingual support for regional farmer accessibility

---

🔮 Future Improvements

• Firebase authentication for cross-device sync
• Cloud-based scan history and farm analytics dashboard
• Grad-CAM heatmap overlay — show which leaf region triggered detection
• Model upgrade to EfficientNet-B3 for improved real-world accuracy
• Offline map integration for field-level disease spread tracking
• Support for additional plant species beyond current 14
• Edge AI optimization for UAV onboard inference

---

👩‍💻 Author

Sneha B.Tech — Electronics and Computer Science (ECS) Project: Krishinoor — Smart Farming · UAV · AI Platform

🔗 ML Model Repository: krishinoor-ml-model

---

📄 License

This project is for educational and research purposes.