PortfolioAI/ML Institute
Back to projects
Project 2
🌾

Agrease

AI for Agricultural Resilience

“Bringing computer vision to the farmland”

TensorFlow Lite
Flutter

Agrease is a mobile application that detects paddy pests and rice leaf diseases using deep learning, helping farmers prevent the 25% of rice crop failures caused by plant diseases. Aligned with SDG #2 Zero Hunger, developed for Google Solution Challenge 2023, achieving Top 100 globally.

🌍 SDG #2 Zero Hunger: Addressing 25% rice crop failures from leaf disease
🎯 95% algorithm accuracy for rice disease detection
📴 Offline mode: Full functionality without internet
🏆 Top 100 globally in Google Solution Challenge 2023
SDG #2 Zero Hunger

SDG #2 Zero Hunger

Team
AJ

Alfiatun Jatnika

Researcher · Biology

MZ

Muhammad Zhafran (Me)

Mobile Dev & MLOps · Computer Science

RH

Ramadhania Humaira

Project Manager · Computer Science

SA

Salman Al Farisi Harahap

ML Engineer · Electrical Engineering

Project Video
25% of rice crop failures occur due to rice leaf disease.

Rice is the staple food for billions, yet a quarter of harvests are lost to preventable leaf diseases. Aligned with SDG #2 Zero Hunger, Agrease puts AI-powered disease detection directly in farmers' hands—no internet required, no expert needed.

Impact

Top 100 globally in Google Solution Challenge 2023. 95% detection accuracy. Offline-first design ensuring accessibility in rural areas without connectivity.

Challenge

25% of rice crop failures occur due to rice leaf disease, threatening food security

Rice farmers lose a quarter of their crops to leaf diseases they can't identify in time. Traditional diagnosis requires expert agronomists who are scarce and expensive. By the time disease is visually detected, it's often too late.

ML Opportunity

An app that detects paddy pests with 95% accuracy, providing prevention advice, treatment recommendations, and product suggestions—all working offline in the field.

Methodology

MLOps pipeline for converting trained models to TFLite and integrating with Flutter mobile app

1
Model Conversion: Converted trained CNN models (MobileNetV2) from SavedModel format to TensorFlow Lite using TFLiteConverter
2
Quantization: Applied post-training quantization to reduce model size from 14MB to ~3MB while maintaining accuracy
3
Flutter Integration: Integrated tflite_flutter package to run TFLite interpreter for on-device inference
4
Camera Pipeline: Built image preprocessing pipeline: camera capture → resize → normalize → TFLite input tensor

Architecture

Offline-first Flutter app with embedded TFLite models for on-device disease classification

Mobile App (Flutter)

Camera Integration
Image Preprocessing
TFLite Interpreter
Results UI
Disease Info Database

ML Models (TFLite)

MobileNetV2 Backbone
38-Class Classification Head
Quantized Weights (.tflite)

Backend Services

Firebase Authentication
Cloud Storage
Diagnosis History Sync

Model Evaluation

Algorithm Accuracy95%

rice disease detection

Inference Time<100ms

on mid-range Android device

Model Size~3MB

after quantization

Offline Support100%

full functionality without internet

Deployment

Bundle TFLite model with Flutter app for immediate offline use

Model Bundling: TFLite model included in app assets; loaded into interpreter on app startup
Offline Mode: Full disease detection without internet; results cached locally
Cross-Platform: Single Flutter codebase for iOS and Android deployment
Model Updates: New model versions downloadable from Firebase Cloud Storage

Successfully deployed ML-powered rice disease detection to mobile devices with offline capability

Competition

Top 100 worldwide in Google Solution Challenge 2023

Role

Mobile Developer & MLOps Engineer

Pest Detection

95% accurate rice leaf disease identification with probability scores

Recommendations

Provides prevention tips, treatment advice, and product recommendations

Offline Mode

Full functionality without internet—designed for rural areas

Key Achievements

Top 100 (Google Solution Challenge)
95% rice disease detection accuracy
SDG #2 Zero Hunger aligned