Vegetable Classification Using CNNs

This project performs image classification on 15 types of vegetables using three different Convolutional Neural Network (CNN) architectures — VGG16, MobileNetV2, and ResNet50. The entire pipeline, including data loading, model training, evaluation, and visualization, is implemented in a single Kaggle notebook.

---

Notebook

The full project is implemented in a Kaggle notebook:

🔗 Three CNN Architectures to Classify Vegetables

---

Dataset

The dataset is publicly available on Kaggle:

🔗 Vegetable Image Dataset by misrakahmed

Dataset Summary

• Total images: 21,000
• Number of classes: 15
• Image size: 224 × 224
• Format: .jpg
• Images per class: 1,400
• Data split: 70% training, 15% validation, 15% testing

Classes

Bean, Bitter Gourd, Bottle Gourd, Brinjal, Broccoli, Cabbage, Capsicum, Carrot, Cauliflower, Cucumber, Papaya, Potato, Pumpkin, Radish, Tomato

---

Models Used

Three popular CNN architectures were used via transfer learning:

• VGG16
• MobileNetV2
• ResNet50

All models were fine-tuned on the same dataset using a consistent training pipeline to compare performance.

---

Evaluation & Visualization

Each model was evaluated using:

• Classification report
• Confusion matrix
• Accuracy and loss plots
• Prediction visualization

---

Libraries & Tools

• TensorFlow / Keras
• Scikit-learn
• NumPy, Pandas
• Matplotlib, Seaborn

---