CNN Image Classification with EfficientNetB4

This project presents an image classification pipeline using a convolutional neural network (CNN) based on EfficientNetB4 architecture. The model is designed to handle a custom image dataset, leveraging advanced data augmentation and regularization techniques to enhance performance and prevent overfitting.

🏢 Model Architecture

• Base Model: EfficientNetB4 (pre-trained on ImageNet)
• Top Layers: Custom classification head
• Regularization: L1, L2 weight penalties
• Augmentation: Random flip, rotation, zoom
• Training Strategy:
  • EarlyStopping to avoid overfitting
  • ModelCheckpoint to save the best model

📁 Dataset

FER_2013

• Structure:

/train
  /class_1
    image1.jpg
    image2.jpg
  /class_2
    ...
/test
  /class_1
    ...
  /class_2
    ...

• Images are automatically loaded and labeled from directory structure using TensorFlow's ImageDataGenerator.

⚙️ Dependencies

• Python 3.x
• TensorFlow 2.x
• NumPy, Pandas, Matplotlib, Seaborn
• scikit-learn

📊 Evaluation

The model's performance is assessed using:

• Confusion Matrix
• Classification Report (Precision, Recall, F1-score)
• Accuracy metrics over training epochs

📈 Visualization

• Loss and Accuracy plots
• Confusion matrix heatmap
• Example predictions

🚀 How to Run

1. Clone this repository:

   git clone https://github.com/your-username/your-repo-name.git
   cd your-repo-name

2. Install the required packages:

   pip install -r requirements.txt

3. Place your dataset inside the train and test directories.

4. Run the notebook:

   jupyter notebook CNN_Model.ipynb

📌 Notes

• Make sure the dataset is well-organized as described.
• You can easily swap EfficientNetB4 with other architectures using Keras Applications.

📧 Contact

For questions or suggestions, feel free to open an issue or contact the maintainer.

---

Author: Farshad Tofighi (farshad257)
License: MIT Email: farshadtfgh@gamil.com