Math 156 Final Project

This repository contains the code and documentation for the Math 156 Final Project, focusing on using Convolutional Neural Networks (CNNs) for car image classification.

Overview

The goal of this project is to classify car images into one of seven classes using a CNN. This classification task addresses real-world applications such as vehicle identification in surveillance systems. The project involves building a CNN model from scratch, training it on a labeled dataset, and evaluating its performance.

Dataset

The dataset used is the Car Images Dataset. It consists of 4,165 images across 7 car classes:

• Audi
• Hyundai Creta
• Mahindra Scorpio
• Rolls Royce
• Swift
• Tata Safari
• Toyota Innova

Preprocessing Steps

• Images were resized to (128 \times 128) pixels.
• Pixel values were normalized to the ([0, 1]) range.
• An 80/20 train-test split was applied.

Model Architecture

The CNN model was built using TensorFlow and includes:

• Convolutional Layers: Extract features from input images.
• Pooling Layers: Reduce spatial dimensions while retaining important features.
• Dropout Layer: Prevent overfitting.
• Fully Connected Layer: Perform the final classification with a softmax activation function.

Key Training Parameters

• Loss Function: Categorical Cross-Entropy
• Optimizer: Adam
• Epochs: 10
• Batch Size: 32

Results

The model achieved the following:

• Training Accuracy: 95.78%
• Validation Accuracy: 67.11%

Observations

• Overfitting was observed after epoch 6.
• Class imbalance affected performance, with some classes like Rolls Royce having lower accuracy.

Future Improvements

• Use balanced datasets to address class imbalance.
• Implement early stopping to prevent overfitting.
• Explore more advanced architectures like ResNet or transfer learning.
• Apply additional data augmentation techniques.

Main References

• Kaggle Dataset: Car Images Dataset
• TensorFlow Documentation: Convolutional Neural Networks

---