Skip to content

Latest commit

 

History

History

image-classification-of-CIFAR-10-using-CNN

README.md

CIFAR-10 Image Classification Using Convolutional Neural Networks (CNNs)

This project involves building and training a Convolutional Neural Network (CNN) to classify images in the CIFAR-10 dataset into 10 classes. The implementation includes data augmentation, hyperparameter tuning, and the use of advanced deep learning techniques to optimize performance.

🎯 Objective

  • Develop a CNN model to classify images in the CIFAR-10 dataset with high accuracy.
  • Understand the impact of hyperparameters like filter size, stride, and dropout on model performance.
  • Apply data augmentation techniques to improve generalization and prevent overfitting.

📂 Dataset

CIFAR-10 Dataset

  • Description: The CIFAR-10 dataset consists of 60,000 32x32 color images in 10 classes, with 6,000 images per class. It is widely used for image classification tasks.
  • Classes: Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship, Truck.
  • Dataset Source: CIFAR-10 Dataset
  • Preprocessing:
    • Normalized pixel values to range [0, 1].
    • Data augmentation applied, including rotation, rescaling, zoom, and horizontal flipping.

🛠️ Tools and Libraries Used

  • TensorFlow & Keras: For building, training, and evaluating the CNN model.
  • Pandas & NumPy: For data manipulation and preprocessing.
  • Matplotlib & Seaborn: For visualizing training metrics and results.

🌟 Model Architecture

The CNN model consists of the following layers:

  1. First Convolutional Block:

    • Convolutional Layer: 32 filters, 3 × 3 kernel, stride 1, "same" padding, "He normal" initialization, ReLU activation.
    • Batch Normalization
    • Dropout (0.2)

  2. Second Convolutional Block:

    • Convolutional Layer: 64 filters, 3 × 3 kernel, stride 1, "same" padding, "He normal" initialization, ReLU activation.
    • Batch Normalization
    • Max Pooling (2 × 2)

  3. Third Convolutional Block:

    • Convolutional Layer: 128 filters, 5 × 5 kernel, stride 1, "same" padding, "He normal" initialization, ReLU activation.
    • Batch Normalization
    • Dropout (0.2)

  4. Fourth Convolutional Block:

    • Convolutional Layer: 256 filters, 3 × 3 kernel, stride 1, "same" padding, "He normal" initialization, ReLU activation.
    • Batch Normalization
    • Max Pooling (2 × 2)

  5. Output Layer:

    • Global Average Pooling
    • Dropout (0.2)
    • Dense Layer with Softmax activation (10 classes)

🚀 Training Details

  • Optimizer: Adam with an initial learning rate of 0.01.
  • Learning Rate Scheduler: ReduceLROnPlateau to reduce learning rate when validation accuracy plateaus (factor 0.5, patience 5, minimum learning rate 0.00001).
  • Loss Function: Categorical Crossentropy.
  • Batch Size: 32
  • Epochs: 50
  • Validation Split: 20% of the training data.
  • Data Augmentation:
    • Rotation (20 degrees)
    • Rescaling (0-1)
    • Zoom (range 0.1)
    • Horizontal flipping

🎉 Results

  • Baseline Model Test Accuracy: ~80%-90%
  • Insights:
    • Larger filter sizes (e.g., 9 × 9) and higher strides (e.g., 2) reduce accuracy due to loss of fine-grained spatial details.
    • Data augmentation improved model generalization and reduced overfitting.

📊 Evaluation Metrics

  • Accuracy
  • Precision, Recall, F1-Score (if applicable)
  • Training and validation loss and accuracy trends.

📜 License

This project is licensed under the MIT License. You are free to use the code and data for educational and non-commercial purposes.

🌐 Connect with Me

Feel free to connect, collaborate, or provide feedback: