This project integrates Deep Learning (DL) and IoT-based soil monitoring to support precision agriculture.
It enables automated detection of rice leaf diseases using image classification models, along with real-time soil health monitoring (NPK, pH, temperature, and moisture).
By combining computer vision with IoT sensors, the system helps farmers with:
- Early detection of rice diseases
- Real-time soil nutrient assessment
- Data-driven recommendations for fertilizer & pesticide use
- A web dashboard for monitoring and visualization
- ✅ Rice Leaf Disease Detection (10 classes, 97.5% accuracy)
- ✅ Soil Health Monitoring using ESP32 + NPK, pH, DHT22 & moisture sensors
- ✅ Deep Learning Models: EfficientNet-B4, DenseNet121, Xception, MobileNetV3, VGG19, InceptionV3
- ✅ Web Application (Flask + Jinja) for real-time disease & soil reports
- ✅ Dashboard with disease predictions, soil nutrient reports, and recommendations
- ✅ Hardware Integration with IoT sensors and ESP32
- Total Images: 11,420 rice leaf images
- Classes (10):
- Bacterial Blight
- Brown Spot
- Leaf Smut
- Leaf Blast
- Tungro
- Sheath Blight
- Bacterial Leaf Streak
- Hispa
- Bacterial Panicle Blight
- Healthy Leaf
- Format: JPEG, size 480×640 px
- Frontend: Jinja web app (dashboard)
- Backend: Flask REST API (model deployment + database)
- IoT Hardware: ESP32 DevKit + Soil Sensors (NPK, pH, Moisture, DHT22)
- Outputs:
- Disease type + confidence score
- Soil nutrient analysis & recommendations
Figure 1: Hardware Setup
Figure 2: Full Project Setup
- Data Preprocessing: Augmentation (flip, rotate, crop, color jitter, Gaussian blur)
- Models: Transfer Learning (EfficientNet, DenseNet, Xception, MobileNet, VGG19)
- Ensemble Model achieved 97.5% accuracy, F1=0.98
- Evaluation Metrics: Accuracy, Precision, Recall, F1-Score, Confusion Matrix
- ESP32 DevKit V1
- NPK Sensor (RS485)
- Capacitive Soil Moisture Sensor
- Soil pH Sensor
- DHT22 (temperature & humidity)
- Power Supply (18650 Li-ion + TP4056 charger)
| Model | Accuracy (%) | F1 Score | Precision | Recall | Loss | Std. Dev. |
|---|---|---|---|---|---|---|
| EfficientNet-B4 | 97.13 | 0.97 | 0.97 | 0.97 | 0.1511 | 0.01 |
| Xception41 | 97.07 | 0.96 | 0.97 | 0.96 | 0.1406 | 0.02 |
| DenseNet121 | 97.20 | 0.98 | 0.99 | 0.97 | 0.1367 | 0.02 |
| DenseNet201 | 97.20 | 0.99 | 0.97 | 0.98 | 0.1312 | 0.02 |
| MobileNetV3-Large | 96.73 | 0.96 | 0.96 | 0.96 | 0.1510 | 0.03 |
| InceptionV3 | 96.53 | 0.96 | 0.96 | 0.96 | 0.1487 | 0.03 |
| AlexNet | 95.60 | 0.95 | 0.95 | 0.95 | 0.2162 | 0.04 |
| ResNet50 | 95.80 | 0.95 | 0.96 | 0.95 | 0.1566 | 0.02 |
| GoogLeNet / Inception-v1 | 97.07 | 0.97 | 0.97 | 0.97 | 0.1267 | 0.02 |
| VGG19 | 93.87 | 0.94 | 0.94 | 0.94 | 0.2907 | 0.03 |
| ShuffleNet | 92.13 | 0.92 | 0.92 | 0.92 | 0.2750 | 0.05 |
| SqueezeNet | 89.73 | 0.90 | 0.90 | 0.90 | 0.3707 | 0.05 |
| Improved LeNet-5 | 86.67 | 0.87 | 0.87 | 0.87 | 0.6007 | 0.06 |
| Ensemble Model | 97.50 | 0.98 | 0.98 | 0.98 | 0.5762 | 0.00 |
Watch the full project demo video here:
Click the thumbnail to watch on YouTube
Estimated project cost: 28,440 – 52,500 BDT
- Hardware (ESP32, sensors, batteries)
- Cloud GPU access for model training
- Deployment & maintenance
- Develop a mobile app for field deployment
- Integrate drone-based imaging for large-scale monitoring
- Expand dataset for better generalization
- Optimize for lightweight edge devices
- Chayon Kumar Das
- Suvro Kumar Das
- Supervised by: Md Toukir Ahmed, Assistant Professor, Dept. of IoT & Robotics Engineering, University of Frontier Technology, Bangladesh
This project is developed for academic and research purposes.
Please cite the authors when using this work.
### 🚀 Getting Started
# Clone repository
git clone https://github.com/shuv001/Capstone-Project-RiceDiseaseDetectionSoilHealthMonitoring.git
cd Capstone-Project-RiceDiseaseDetectionSoilHealthMonitoring
# Install dependencies
pip install -r requirements.txt
# Run the web app
python app.py