Image Classification with CNN

Image Classification with CNN

This project focuses on classifying images from the CIFAR-10 dataset using Convolutional Neural Networks (CNNs). The goal was to improve baseline accuracy through optimized model architecture, data preprocessing, and experimentation with different deep learning models.

Built as part of the AI Engineering Bootcamp (Saudi Digital Academy), the system explores multiple state-of-the-art CNN architectures and demonstrates practical skills in model design, training, evaluation, and fine-tuning.

Objectives:

Build and train deep learning models for image classification tasks.

Improve accuracy over baseline models using advanced CNN architectures.

Compare the performance of different models (ResNet, VGG16, InceptionNet, EfficientNet).

Optimize preprocessing techniques to enhance model generalization.

Key Features:

Data Preprocessing

Applied normalization, resizing, and augmentation (flipping, rotation, cropping).

Improved model generalization by increasing dataset diversity.

CNN Model Development

Built a custom CNN and trained it on CIFAR-10.

Implemented and compared performance across pretrained models: ResNet, VGG16, InceptionNet, and EfficientNet.

Performance Optimization

Achieved 92% accuracy and F1 score, surpassing the 85% baseline.

Tuned hyperparameters including learning rate, batch size, and dropout.

Evaluation Metrics

Evaluated models using accuracy, precision, recall, F1 score, and confusion matrix.

Visualized learning curves and prediction results for analysis.

List of tools and technologies:

Programming Language: Python

Deep Learning Libraries: PyTorch, TensorFlow, Keras

Image Processing: OpenCV, torchvision, Pillow

CNN Architectures: Custom CNN, ResNet, VGG16, InceptionNet, EfficientNet

Data Handling: pandas, NumPy

Visualization: Matplotlib, Seaborn

Model Evaluation: scikit-learn (for metrics and confusion matrix)

Version Control: Git, GitHub

Training Environment: Google Colab, Jupyter Notebook