🍅 Tomato Plant Disease Detection using Deep Learning

This project detects diseases in tomato plant leaves using a Convolutional Neural Network (CNN). It uses a Kaggle dataset and provides a user-friendly web interface built with Streamlit. The application can also be deployed using Docker for easy sharing and demonstration.

---

📌 Features

• Tomato leaf disease classification (11 classes)
• CNN model trained using TensorFlow / Keras
• Upload leaf images and get real-time predictions
• Streamlit-based interactive GUI
• Dockerized for consistent deployment
• Clean GitHub repository structure with screenshots and documentation

---

📂 Project Structure

tomato-disease-detection/ |

├── app.py # Streamlit application

├── Dockerfile # Docker deployment

├── requirements.txt # Python dependencies

├── README.md # Project documentation

├── tomato_model.h5 # Trained CNN model

├── src/ # Training scripts and preprocessing

│ ├── train.py

│ ├── model.py

│ └── preprocess.py

├── screenshots/ # Screenshots for README & PPT

---

📊 Dataset

• Source: Kaggle – Tomato Plant Disease Dataset
• Total Classes: 11 (healthy + diseased)
• Dataset is not included due to size constraints

Classes: Bacterial_spot, Early_blight, Late_blight, Leaf_Mold, Septoria_leaf_spot, Spider_mites Two-spotted_spider_mite, Target_Spot, Tomato_Yellow_Leaf_Curl_Virus, Tomato_mosaic_virus, healthy, powdery_mildew

---

🧠 Model Details

• Model Type: Convolutional Neural Network (CNN)
• Framework: TensorFlow / Keras
• Image Size: 224 × 224
• Optimizer: Adam
• Loss Function: Categorical Crossentropy
• Epochs: 10
• Batch Size: 32
• Validation Accuracy: ~78–82%

---

📈 Epoch-wise Training Summary

Epoch	Training Accuracy	Validation Accuracy	Training Loss	Validation Loss
1	42%	49%	1.61	1.52
2	60%	66%	1.10	0.94
3	65%	74%	0.97	0.75
4	69%	71%	0.87	0.93
5	71%	76%	0.81	0.67
6	73%	75%	0.76	0.84
7	73%	77%	0.74	0.71
8	75%	78%	0.68	-
9	76%	79%	-	-
10	77–80%	78–80%	-	-

Observations:

• Steady increase in accuracy shows effective learning.
• Validation accuracy closely tracks training accuracy → good generalization.
• Minor fluctuations in validation loss are expected due to dataset diversity.
• CPU-only training caused longer epoch durations (~30–40 min per epoch).

Limitations:

• Accuracy can improve with transfer learning (MobileNetV2, ResNet50) and GPU acceleration.
• Model is trained on a fixed dataset and may need retraining for real-world scenarios.

---

🖼️ Screenshots

• Model Training
• Streamlit Interface
• Prediction Result

---

🚀 How to Run

Setup (Without Docker)

git clone https://github.com/DudalaShrujana/tomato-disease-detection.git
cd tomato-disease-detection
pip install -r requirements.txt
streamlit run app.py

Open browser: http://localhost:8501

Docker Deployment

# Build Docker Image
docker build -t tomato-disease-app .

# Run Container
docker run -d -p 8501:8501 tomato-disease-app

Open browser: http://localhost:8501

---

📦 Future Scope

• Improve accuracy using transfer learning (MobileNetV2, ResNet50)
• GPU acceleration for faster training
• Web interface improvements (image gallery, batch prediction)
• Extend to other crops and plant diseases