Dog Breed Classification with Convolutional Neural Network (AlexNet, GoogLeNet and MobileNet V2)

Chia, Kit Yau (2023) Dog Breed Classification with Convolutional Neural Network (AlexNet, GoogLeNet and MobileNet V2). Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RDS_Chia Kit Yau_FullText.pdf Restricted to Registered users only Download (2MB)

Abstract

The purpose of this project is to train the 3 convolutional neural networks to classify the different types of dog breeds. By using bare human eyes, we aren't able to identify all of the dog breeds around the world, even with the professionals. We propose this solution to help to identify the breed of the dog fastly and accurately. This could be beneficial to the users such as the pet owners, veterinarians, pet insurance companies and also the government. This project covers the area of Artificial Intelligence, Machine Learning, Deep Learning, Computer Science, Computer Vision and Image Processing. Throughout this project, convolutional neural networks (CNN) will be the deep learning algorithm to allow us to conduct this project. The 3 convolutional network models which are AlexNet, GoogLeNet, and MobileNet. Jupyter Notebook and Google Collab will serve as an computing platform to run the algorithms. Besides that, OpenCV, Tensorflow and PyTorch were used as the main computer vision libraries to execute the models. CUDA from NVIDIA would help to shorten the time needed for training the models. Training and testing data were splitted before the model fitting. The testing data will be treated as the validation dataset for the trained models. Fine Tuning will be carried out before the model fitting. The accuracy as well as the performance will be collected and to be compared on which is the best model. By choosing the best model, we undergo the ensemble model process to achieve a better performance of the model.