🌱 AgroSmart – Smart Irrigation System for Hilly Regions

AgroSmart is an IoT and Machine Learning–based smart irrigation system designed specifically for hilly and water-scarce regions, where irrigation is challenging due to uneven terrain, limited connectivity, and inefficient water usage.

The system combines real-time sensor data, edge-based machine learning, and automated actuation to deliver precise irrigation decisions with high accuracy.

---

📌 Project Overview

AgroSmart predicts whether irrigation is required based on environmental and soil conditions using a Random Forest machine learning model deployed directly on an ESP32-S3 microcontroller.

The system:

• Reduces water wastage
• Prevents over- and under-irrigation
• Works reliably in low-connectivity rural areas
• Is optimized for real-world agricultural constraints

---

🚀 Key Features

• 93% irrigation prediction accuracy using Random Forest
• Edge ML inference on ESP32-S3 (no cloud dependency)
• Real-time decision-making (<30 ms inference time)
• Multilingual, farmer-friendly web dashboard
• Water tank level–aware safety logic
• Optimized for hilly terrain conditions
• Scalable architecture with optional cloud integration

---

🧠 Machine Learning Approach

Problem Type

Binary classification:

• 1 → Irrigation Needed
• 0 → No Irrigation

Model Used

• Random Forest Classifier
• Lightweight and optimized for embedded deployment

Input Features (12 Total)

Base Sensor Features:

• Soil Moisture (%)
• Temperature (°C)
• Humidity (%)
• Water Level (%)

Engineered Features:

• Evapotranspiration rate
• Soil–temperature stress
• Moisture deficit
• Combined stress index
• Vapor Pressure Deficit (VPD) index
• Temperature–humidity ratio
• Critical dry flag
• Optimal moisture flag

---

📊 Model Performance

• Accuracy: ~93%
• Balanced Accuracy: ~93%
• False Negative Rate: ~6%
• False Positive Rate: ~7%
• Model Size: ~40 KB
• Inference Time: <30 ms
• Suitable for ESP32 SRAM constraints

---

⚙️ System Workflow

1. Sensors collect real-time data (soil moisture, temperature, humidity, water level)
2. Feature engineering is performed on-device
3. Random Forest model predicts irrigation requirement
4. Safety checks validate water availability
5. Relay-controlled pump is activated if required
6. Data is optionally sent to a web dashboard

---

🛠️ Hardware Components

• ESP32-S3 microcontroller
• Capacitive soil moisture sensor
• DHT11 temperature & humidity sensor
• Water level sensor
• Relay module
• Water pump / solenoid valve

---

💻 Software Stack

• Python (model training & deployment tools)
• Scikit-learn (Random Forest)
• Flask (web dashboard & API)
• C/C++ (ESP32 deployment)

---

▶️ Getting Started

1. Install Python dependencies: pip install -r requirements.txt

2. Train the ML model: python model_train.py

3. Generate ESP32 deployment code: python deploy.py

4. Flash generated code to ESP32-S3

5. Start the web dashboard: python app.py

---

🏆 Achievements & Validation

• Top 45 teams at VIIT Internal Hackathon Round, Smart India Hackathon (SIH) 2025
• Presented at a Springer-indexed international conference
• Showcased at KISAN AgriShow 2025, Pune
• Strong validation from farmers and agri-professionals

---

👥 Team

Team M.A.R.S

• Manas Kulkarni
• Atharva Maslekar
• Atharva Suryavanshi
• Atharva Rajendra Joshi
• Rajlakshmi Desai
• Samiksha Nalawade

Guided by:

• Dr. Pravin Gawande
• Dr. Snehal Rathi
• Rushikesh Tanksale

---

🎯 Conclusion

AgroSmart demonstrates how IoT, machine learning, and embedded systems can be combined to build a practical, scalable, and farmer-centric irrigation solution. The project moves beyond theory into real-world deployment, validation, and impact—especially for challenging hilly regions.