Car Model Categorization with Different Kind of Deep Learning Convolutional Neural Network Models
AbstractWith the increasing of data size in recent years, processing and analyzing of the big data has become difficult. Therefore, interest in this area has also gone up. In 2006, Geoffrey Hinton, the pioneers in the field of machine learning, developed a deep learning model at the University of Toronto. The basis of the deep learning model is based on the use of a large number of hidden layers in artificial neural networks. Previously, increasing the number of hidden layers in artificial neural networks, deepening the network, was not preferred because it caused complexity in calculation processes. Deep learning that provides a nonlinear transformation of the big data can model complex relationships with a multi-layered structure. As this model provides more comprehensive attribute learning, the classification becomes more successful. For this study, Convolutional Neural Network (CNN) model is represented for the large-scale image classification. CNN is the most widely used deep learning model in feature learning, recognition, and classification. In order to obtain a powerful image classifier, there should be large amount training data. Because of that, data augmentation techniques are used to boost the performance of the deep networks. In this study, data was increased using image augmentation methods that create artificial training examples through different ways of processing techniques such as adding noise, rotating, shifting, shearing, and resizing. The data that consists of car photos taken from the rear and are provided from the literature are separated into 40 categories. Finally, the classification performances using different deep learning CNN models such as AlexNet, VGG16, and VGG19 are evaluated and experimental results are reported.