🎭 Facial Emotion Detection using Deep Learning

📋 Overview

A deep learning project implementing multiple CNN architectures to detect four emotional states (happy, sad, surprised, neutral) from facial expressions. Achieved 84.38% accuracy using transfer learning approaches.

⭐ Key Features

• 11 different CNN architectures implemented
• Transfer learning with VGG16, ResNet V2, and EfficientNet
• Custom CNN with Residual and SE blocks
• Both grayscale and RGB image processing

🏆 Best Models

1. VGG16 Transfer Learning: 84.38% accuracy
2. Complex Custom CNN: 82.81% accuracy
3. Deep CNN (Grayscale): 80.47% accuracy

💻 Technologies

• TensorFlow/Keras
• Python
• NumPy
• Matplotlib
• OpenCV

📊 Dataset

• ~15,000 training images
• 48x48 pixel images
• Four emotion classes
• Both grayscale and RGB formats

📈 Results

• Best F1-score: 0.85 (VGG16 model)
• Strong performance across all emotion classes
• Effective handling of class imbalance

👨‍💻 Author

• Ankit Rai

This project demonstrates the power of deep learning in emotion recognition through facial expression analysis.