Comparative Study of Convolutional Neural Networks-based Algorithm for Fine-grained Car Recognition

Sanjaya, Joseph and Ayub, Mewati and Toba, Hapnes (2021) Comparative Study of Convolutional Neural Networks-based Algorithm for Fine-grained Car Recognition. In: 1st International Conference on Emerging Issues in Technology, Engineering and Science, Bandung, July 1-2, 2021.

	Text 2. 2021-Comparative Study of Convolutional Neural Networks-based Algorithm for Fine-grained Car Recognition (1).pdf Download (42Mb)
	Text 10_ Comparative Study of Convolutional Neural Networks-based Algorithm for Fine-grained Car Recognition.pdf Download (2493Kb)

Abstract

The use of the Deep-Learning model for object recognition in vision machines has been widely applied. Convolutional Neural Network (CNN) is one of the algorithms which has achieved a significant progress in object recognition task. An algorithm that has good accuracy and speed is required to recognize a car specification. This research presents a comparative study of several CNN models for car recognition. This study is a continuation of previous study about data augmentation in car image recognition using ResNet architecture. In this study, the CNN architectures which are used in comparison, are ResNet, SqueezeNet, and EfficientNet. The aim of this study is to find an architecture with optimal performance in car recognition. The dataset used is a Cars Dataset provided by Stanford University. The methods consist of data pre-processing, model training and hyper parameter tuning, inferences and comparison. The metrics which were used during the experiments are accuracy, model size, and speed. Training of each model was performed using computer with the same specification. The experimental results indicate that EfficientNet model gives the best result among other models in the context of accuracy, model size, and speed.

Item Type:

Conference or Workshop Item (Paper)

Contributors: