Machine-Learning-II-Project

Object detection and Classification using Convolutional Neural Networks on CIFAR-10

Overview

This project explores the impact of different hyperparameters on Convolutional Neural Networks (CNNs) using the CIFAR-10 dataset. Our goal is to optimize the model architecture and preprocessing techniques to enhance the accuracy and efficiency of CNNs in image classification tasks.

Project Objectives

1. Hyperparameter Optimization: Explore various hyperparameter configurations to find the best settings for improving model accuracy and efficiency.
2. Data Preprocessing and Enhancement: Use advanced preprocessing and data augmentation techniques to prepare and enhance the dataset for building a robust CNN model.

Dataset

CIFAR-10

The CIFAR-10 dataset consists of 60,000 32x32 color images in 10 classes, with 6,000 images per class. There are 50,000 training images and 10,000 testing images.

Directory Structure

• data/
  • train/
  • validation/
  • test/
• models/
  • model_1.h5
  • model_2.h5
  • model_3.h5
  • model_4.h5
  • model_5.h5
  • model_6.h5
  • model_7.h5
  • model_8.h5
• notebooks/
  • model_1.ipynb
  • model_2.ipynb
  • model_3.ipynb
  • model_4.ipynb
  • model_5.ipynb
  • model_6.ipynb
  • model_7.ipynb
  • model_8.ipynb
• README.md

Data Preprocessing

1. Normalization: Pixel values are scaled to the range [0, 1].
2. One-Hot Encoding: Class labels are converted into binary vectors using tf.keras.utils.to_categorical().
3. Data Augmentation: Techniques like rotation, zooming, and shifting are applied to increase training data variety and model robustness.

Model Architecture

General CNN Structure

1. Input Layer: 32x32 RGB images.
2. Convolutional Layers: Apply filters to extract features.
3. Activation Functions: Introduce non-linearity (ReLU).
4. Pooling Layers: Reduce spatial dimensions (Max Pooling).
5. Flatten Layer: Converts 2D feature maps into a 1D vector.
6. Dense (Fully Connected) Layers: Aggregate features to make predictions.
7. Output Layer: Softmax activation for classification.

Detailed Architecture for Each Model

The project includes eight different models, each with unique configurations of convolutional layers, pooling layers, dense layers, activation functions, and regularization techniques.

Model 1

• 2 Convolutional Layers
• Max Pooling
• Flatten
• 1 Dense Layer
• Softmax Output

Model 2

• 3 Convolutional Layers
• Max Pooling
• Flatten
• 2 Dense Layers
• Softmax Output

Model 3

• 4 Convolutional Layers
• Max Pooling
• Flatten
• 2 Dense Layers
• Softmax Output

Model 4

• 5 Convolutional Layers
• Max Pooling
• Flatten
• 3 Dense Layers
• Softmax Output

Model 5

• Batch Normalization after each Convolutional Layer
• 3 Convolutional Layers
• Max Pooling
• Flatten
• 2 Dense Layers
• Softmax Output

Model 6

• Dropout after each Dense Layer
• 4 Convolutional Layers
• Max Pooling
• Flatten
• 2 Dense Layers
• Softmax Output

Model 7

• Increased Filter Size in Convolutional Layers
• 3 Convolutional Layers
• Max Pooling
• Flatten
• 2 Dense Layers
• Softmax Output

Model 8

• Combination of Batch Normalization and Dropout
• 4 Convolutional Layers
• Max Pooling
• Flatten
• 3 Dense Layers
• Softmax Output

Training and Evaluation

Compilation

• Optimizer: Adam with a learning rate of 0.001
• Loss Function: Categorical cross-entropy
• Metrics: Accuracy

Training

Models are trained using the fit() function with:

• Batch Size: Number of samples processed before updating weights.
• Number of Epochs: Total iterations over the dataset.
• EarlyStopping: To halt training if performance on the validation set doesn’t improve.

Evaluation

• Accuracy and Loss: Measure learning and generalization.
• Confusion Matrix: To understand the classification performance.
• Training and Validation Curves: Analyze accuracy and loss curves for insights.

Tools and Libraries

• Keras: High-level neural network API for fast prototyping.
• TensorFlow: Open-source library for machine learning and deep learning.
• Matplotlib: For plotting training and validation metrics.

Conclusion

This project aims to optimize CNN models for better image classification using the CIFAR-10 dataset. By fine-tuning hyperparameters and applying advanced preprocessing techniques, we developed CNN algorithms that excel in accuracy and generalization.

How to Use

1. Clone the repository:

   git clone https://github.com/karim7tr/Machine-Learning-II-Project.git
   cd Machine-Learning-II-Project

2. Install required libraries:

   pip install -r requirements.txt

3. Run the Jupyter notebooks:

   jupyter notebook notebooks/model_1.ipynb

4. Train the models: Follow the instructions in each notebook to train and evaluate the models.

teamMates.

• Karim Triki
• Ines Haouala