Capstone Final Project

• Titweng: Cattle Identification Framework for Mitigating Raiding and Enhancing Market Trust in South Sudan
• Biometric nose print recognition system that prevents cattle theft, resolves ownership disputes, and builds trust in livestock markets through secure digital identification.

Project Overview

Problem Statement

South Sudan's livestock sector faces endemic cattle raiding and ownership disputes due to the absence of reliable cattle identification systems. Traditional methods and oral testimony create verification failures, allowing stolen cattle to enter markets with impunity and perpetuating cycles of conflict and economic loss.

Solution

Titweng is a comprehensive digital cattle identification framework designed to mitigate raiding and enhance market trust in South Sudan. By leveraging biometric nose print recognition - as unique as human fingerprints - the system creates tamper-proof digital identities for cattle, enabling secure ownership verification and building confidence in livestock markets. This framework transforms cattle from untraceable assets into digitally verified commodities, making theft unprofitable and disputes resolvable through objective evidence.

Key Features

• Biometric Registration: Capture and register cattle nose prints with 93.4% accuracy
• Real-time Verification: Instant cattle identification in under 3 seconds
• Cross-platform Access: Mobile app (Android/iOS) and web dashboard
• Digital Certificates: QR-coded ownership certificates with PDF generation
• Secure Database: PostgreSQL with vector embeddings for scalable matching
• Automated Notifications: Email/SMS alerts for registrations and transfers
• Analytics Dashboard: Real-time monitoring and reporting for administrators

Demo Video

Final Product Demo Video

Screenshot Highlights

Live Deployment

• Admin Dashboard: https://www.titweng.com
• Mobile APK: Download Android App

SMS, Email, Reciept Screenshots

Technology Stack

• Mobile App: Flutter 3.0+ (Android/iOS)
• Admin Dashboard: React 18 + TypeScript
• Backend API: FastAPI + Python 3.8+
• Database: PostgreSQL + pgvector
• ML Models: YOLOv8 + Siamese CNN
• Deployment: Netlify + Azure

Installation & Setup

Prerequisites

• Python 3.8+, Node.js 16+, Flutter 3.0+, PostgreSQL 12+

1. Backend Setup

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Environment variables (.env file)
DATABASE_URL=postgresql://user:password@localhost:5432/titweng
ADMIN_USERNAME=admin
ADMIN_PASSWORD=secure_password
BREVO_API_KEY=your-brevo-key

# Start server
uvicorn main:app --reload --host 0.0.0.0 --port 8000

2. Admin Dashboard Setup

cd admin_dashboard
npm install
npm run dev  # Development
npm run build  # Production

3. Mobile App Setup

cd mobile_app
flutter pub get
flutter run  # Development
flutter build apk --release  # Production APK

4. Database Setup

CREATE DATABASE titweng;
CREATE EXTENSION IF NOT EXISTS vector;

Testing Results

Testing Strategy 1: Functional Testing

Test Case	Input Data	Expected Result	Actual Result	Status
Valid Registration	3-5 nose images, owner details	Success + Certificate	Success + PDF generated	PASS
Invalid Images	Blurry/dark images	Error message	"Image quality too low"	PASS
Duplicate Registration	Same cow twice	Duplicate detection	"Similar cow found (85% match)"	PASS

Testing Strategy 2: Performance Testing

Test Case	Concurrent Users	Response Time (avg)	Success Rate	CPU Usage	Memory Usage	Status
Light Load	10 users	1.2s	100%	45%	2.1GB	PASS
Medium Load	50 users	2.8s	98%	78%	3.4GB	PASS
Heavy Load	100 users	4.5s	95%	92%	4.8GB	PASS
Stress Test	150 users	7.2s	89%	98%	6.1GB	DEGRADED
Peak Load	200 users	12.1s	76%	100%	7.8GB	FAIL

Testing Strategy 3: Hardware Performance Testing

Test Case	Device Specification	Registration Time	Verification Time	App Launch Time	Status
Premium Device	iPhone 14 Pro (8GB RAM, A16)	3.2s	1.8s	2.1s	PASS
High-end Android	Samsung S23 (8GB RAM, Snapdragon 8)	3.5s	2.0s	2.4s	PASS
Mid-range Device	Samsung A54 (6GB RAM, Exynos 1380)	5.1s	2.9s	3.4s	PASS
Budget Device	Redmi Note 11 (4GB RAM, Snapdragon 680)	6.8s	3.7s	4.2s	PASS
Low-end Device	Android Go (2GB RAM, Unisoc Tiger)	8.7s	4.2s	5.8s	SLOW

Testing Strategy 4: Accuracy Testing with Different Data Values

Test Case	Data Variation	Sample Size	True Positive Rate	False Positive Rate	F1-Score	Status
Optimal Lighting	Bright daylight conditions	500 images	94.2%	2.1%	0.96	PASS
Low Light	Indoor/evening conditions	300 images	89.7%	4.3%	0.92	PASS
Variable Angles	Front, side, angled shots	300 images	91.8%	3.4%	0.94	PASS
High Quality	4K camera, clear images	400 images	96.1%	1.8%	0.97	PASS
Compressed Images	Low resolution, compressed	400 images	89.5%	4.2%	0.92	PASS
Young Cattle	Calves under 1 year	200 images	87.3%	5.1%	0.91	PASS
Mature Cattle	Adult cattle over 3 years	200 images	92.8%	2.9%	0.95	PASS

Cross-Validation Results: 93.4% ± 0.5% average accuracy across 5 folds

📊 Analysis

Objectives Achievement Analysis

Successfully Achieved Objectives (100%)

1. Biometric Identification System

   • Implemented YOLOv8 for nose detection with 94.2% accuracy
   • Developed Siamese CNN for embedding generation
   • Achieved 93.4% average identification accuracy

2. Multi-platform Application

   • Flutter mobile app supporting Android/iOS
   • React web dashboard for administrators
   • RESTful API backend with comprehensive endpoints

3. Real-time Processing

   • Average verification time: 2.1s on mid-range devices
   • Concurrent user support up to 100 users
   • Real-time database updates and notifications

Partially Achieved Objectives (70%)

1. Offline Functionality
   • Mobile app supports offline image capture
   • Limited offline verification capability
   • Gap: Full offline processing requires optimization

Missed Objectives (0%)

1. Blockchain Integration
   • Reason: Technical complexity exceeded project timeline
   • Impact: Reduced decentralization features
   • Mitigation: Implemented robust database logging instead

Performance Analysis

• Accuracy: 93.4% average identification accuracy exceeds industry standards (85-90%)
• Speed: Sub-3-second verification meets real-time requirements
• Scalability: Successfully handles 100 concurrent users
• Reliability: 98% uptime during testing period

Deployment

Production Deployment Steps

1. Backend Deployment (Azure)

# Install Azure CLI and login
az login
az webapp create --resource-group titweng-rg --plan titweng-plan --name titweng-app
az webapp config appsettings set --name titweng-app --resource-group titweng-rg --settings DATABASE_URL=postgresql://...
az webapp deployment source config-zip --name titweng-app --resource-group titweng-rg --src backend.zip

2. Frontend Deployment (Netlify)

cd admin_dashboard
npm run build
# Deploy to Netlify via drag-and-drop or CLI
# Set environment variables in Netlify dashboard:
# VITE_API_URL=https://titweng-app-a3hufygwcphxhkc2.canadacentral-01.azurewebsites.net

3. Database Setup (Production)

-- Create database with pgvector
CREATE EXTENSION IF NOT EXISTS vector;
-- Tables created automatically on first run
CREATE INDEX ON embeddings USING ivfflat (embedding vector_cosine_ops);

4. Mobile App Distribution

flutter build apk --release --split-per-abi
# Upload to Play Store Console or distribute via GitHub Releases

Deployment Verification

• Health Check: GET /health endpoint for system status
• Performance Benchmarks: API response time < 2s, uptime > 99%
• Monitoring: Error tracking and uptime monitoring configured
• Backup Strategy: Daily automated database backups

Contact

• Project Lead: Geu Aguto Titweng
• Email: g.bior@alustudent.com
• GitHub: @Geu-Pro2023