Abstract
The ensuing project delves into the theory and practical application of Convolutional Neural Networks (CNNs) within the domain of image classification. The application of this study focuses on differentiating between the Sedan, SUV, and Truck vehicle body styles. The theoretical foundation of CNNs is explored, emphasizing relevant machine-learning topics, their architectural components, and how these components interact to enable CNN functionality. The project's practical aspect involves the development, training, and hyperparameter tuning of a convolutional neural network model using a diverse, self-curated dataset comprising different vehicles. Through rigorous experimentation, the final CNN model achieves a satisfactory accuracy of 77\% on the test dataset, showcasing its efficacy in classifying car images into predefined categories. Furthermore, the project utilizes Gradient-weighted Class Activation Mapping (Grad-CAM) to examine how the CNN model makes decisions. This technique enhances interpretability and understanding of the behavior of the model by revealing the areas of the input images that have the greatest influence on the classification decision. Overall, this project serves as a comprehensive investigation of convolutional neural networks, demonstrating their usefulness in image classification tasks and offering insights into their inner workings through sophisticated visualization techniques such as Grad-CAM.
Advisor
Long, Colby
Department
Statistical and Data Sciences
Recommended Citation
Fischer, Paul Thomas, "The Theory and Application of Convolutional Neural Networks for Vehicle Body Style Recognition" (2024). Senior Independent Study Theses. Paper 11073.
https://openworks.wooster.edu/independentstudy/11073
Disciplines
Applied Statistics | Artificial Intelligence and Robotics | Data Science | OS and Networks | Systems Architecture
Keywords
Convolutional Neural Networks, Vehicles, Neural Networks, Machine Learning, Deep Learning, Grad-CAM, Image Recognition, Computer Vision, Body Styles
Publication Date
2024
Degree Granted
Bachelor of Arts
Document Type
Senior Independent Study Thesis
© Copyright 2024 Paul Thomas Fischer
