🍅 A Comparative Analysis of Deep and Traditional Learning Models for Tomato Leaf Disease Classification
Tomato-Leaf is a research project that compares the performance of traditional Machine Learning (ML) models and modern Deep Learning (DL) architectures in classifying tomato leaf diseases. With a focus on agricultural AI, this work explores how different models behave on real-world data, including the impact of background removal on performance.
TOMATO-LEAF/
│
├── dataset/ # Original dataset with disease folders
├── tomato\_bg\_removed/ # Background-removed version of dataset
├── models/ # Saved DL models (.h5 files)
│ ├── best\_inceptionv3.h5
│ └── xception\_tomato\_leaf\_model.h5
├── metrics and visualizations/ # Evaluation visuals
│ ├── confusion\_matrix.png
│ ├── inception\_confusion\_matrix.png
│ ├── inception\_training\_metrics.png
│ └── training\_history.png
├── notebooks/ # All model training notebooks
│ ├── bg\_removed\_inception.ipynb
│ ├── EfficientB1.ipynb
│ ├── inception.ipynb
│ ├── KNN+NB+XGBoost.ipynb
│ ├── Mobile\_shuffle\_efficient.ipynb
│ └── Xception.ipynb
└── readme.md # Project documentation
-
Source: Custom-collected dataset of tomato leaf images.
-
Classes:
- Tomato__Target_Spot
- Tomato__Tomato_mosaic_virus
- Tomato__Tomato_YellowLeaf__Curl_Virus
- Tomato_Bacterial_spot
- Tomato_Early_blight
- Tomato_healthy
- Tomato_Late_blight
- Tomato_Leaf_Mold
- Tomato_Septoria_leaf_spot
- Tomato_Spider_mites_Two_spotted_spider_mite
-
Dataset Variants:
- Original dataset
- Background-removed dataset (via preprocessing)
Implemented in KNN+NB+XGBoost.ipynb:
- K-Nearest Neighbors (KNN)
- Naive Bayes (NB)
- XGBoost
Features were extracted using pre-trained CNNs or hand-crafted descriptors.
| Model | Notebook | Accuracy (%) | Notes |
|---|---|---|---|
| Xception | Xception.ipynb |
97.00 | Best performing overall |
| EfficientNetB0 | EfficientB1.ipynb |
94.26 | Excellent efficiency-accuracy |
| MobileNetV2 | Mobile_shuffle_efficient.ipynb |
92.79 | Lightweight & fast |
| ShuffleNetV2 | Mobile_shuffle_efficient.ipynb |
90.54 | Lightweight, less accurate |
| InceptionV2 | inception.ipynb |
83.00 | Underperformed in this task |
- Accuracy
- Precision, Recall, F1-Score
- Confusion Matrix
- Training/Validation Curves
- Visuals available in
metrics and visualizations/
- Xception gave the highest accuracy (97%), making it ideal for deployment.
- EfficientNetB0 offered a good trade-off between speed and performance.
- MobileNetV2 and ShuffleNetV2 are suitable for resource-constrained environments.
- InceptionV2 was less effective despite its complexity.
- Traditional ML models are faster to train but were outperformed by DL models.
- Removing background boosted model performance by eliminating irrelevant noise.
- Clone the repo:
git clone https://github.com/yourusername/tomato-leaf-disease-classification cd tomato-leaf-disease-classification
2. Install dependencies:
```bash
pip install -r requirements.txt
```
3. Run any Jupyter notebook in `notebooks/`.
---
## 🚀 How to Use
1. Use `KNN+NB+XGBoost.ipynb` for classical ML models.
2. Use the respective `.ipynb` files for DL model training.
3. Pre-trained models are available in the `models/` folder.
4. Evaluate performance using visuals from `metrics and visualizations/`.
---
## 🧠 Future Scope
* Include attention-based architectures (e.g., Vision Transformers)
* Fine-tune models with real-world data augmentation
* Deploy top-performing model into a **web/mobile plant health app**
* Test the model on other plant species and diseases
---
## 👨💻 Author
**Obidur Rahman (Ashfin)**
🔗 [YouTube](https://www.youtube.com/@ObidurRahman) | 💬 Educator, ML Researcher & Content Creator
🧠 Passionate about AI, education, and solving real-world problems with tech.
---
## 📜 License
This project is open-source under the [MIT License](LICENSE).
---
*Made with ❤️ and a lot of tomato leaves!*
```