Vehicle classification under real-world conditions: A comparative analysis of CNN and EfficientNet
DOI:
https://doi.org/10.54097/9jgcpx41Keywords:
Vehicle classification; convolutional neural network; EfficientNet; deep learning.Abstract
This paper uses deep learning techniques for vehicle type classification in real world traffic situations. A dataset of 300 images of cars, trucks, and buses was created. Images were resized to 224x224 pixel dimensions. For deep learning models, images were preprocessed by way of flipping, rotating, and changing brightness to diversify the dataset. For this study, we used compared two deep learning methods: A simple convolutional neural network, and in this case, EfficientNet , which involves transfer learning. During this study, the basic CNN was able to classify images with 68 % accuracy while EfficientNet was able to classify images with 96 % accuracy. This research strongly indicates, and it is to be expected, that the selection of advanced network designs aids in the classification of vehicles. More work needs to be done in optimizing the model’s reliability and practicality, and this can be done by increasing the size of the dataset, adding complexity to the data, and testing with different network designs.
Downloads
References
[1] Yoshua Bengio et al. (2015) Deep Learning p.436–444 Link: https: // www.nature.com/articles/nature14539. DOI: https://doi.org/10.1038/nature14539
[2] Yun Hui Lin, Qingyun Tian (2020) Deep Learning in Intelligent Transportation Systems Section 1 / Introduction Link: https: //arxiv.org/abs/2002.09844.
[3] Benjamin Coifman a, Hunter Danford b, Mo Wu c, Yuyi Chang d, Keith Redmill d (2026) Vehicle Detection and Classification Section 1-2 Link: https :// www. sciencedirect.com/science/ article/pii/ S0968090X17300917. DOI: https://doi.org/10.1080/15472450.2025.2603541
[4] Alex Krizhevsky (2017) ImageNet Classification with Deep CNN p.1097-1105 Link: https://dl.acm.org/doi/10.1145/3065386.
[5] Richard Szeliski (2022) Computer Vision: Algorithms and Applications Section2-3 Link: https://szeliski.org/Book/. DOI: https://doi.org/10.1007/978-3-030-34372-9
[6] Alex Krizhevsky et al. (2017) ImageNet Classification with Deep Convolutional Neural Networks p.1097-1105 Link: https://dl.acm.org/doi/10.1145/3065386. DOI: https://doi.org/10.1145/3065386
[7] Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, Youngjoon Yoo (2019) A Survey on Image Data Augmentation Section2-3 Link: https://arxiv.org/abs/1905.04899.
[8] Yann LeCun (1998) Gradient-Based Learning Applied to Document Recognition p.2278-2324 Link: http://yann.lecun.com/exdb/publis/pdf/lecun-98.pdf. DOI: https://doi.org/10.1109/5.726791
[9] Mingxing Tan (2019) EfficientNet: Rethinking Model Scaling for CNNs Section3-4 Link: https://arxiv.org/abs/1905.11946.
[10] Diederik P. Kingma (2014) Adam: A Method for Stochastic Optimization Section 2 Link: https://arxiv.org/abs/1412.6980.
[11] Adam Falkowski, Giovanni Grilli di Cortona, Zahra Tabrizi (2018) Deep Learning for Visual Recognition: A Review Section3 Link: https://arxiv.org/abs/1802.08296. DOI: https://doi.org/10.1007/JHEP04(2018)101
[12] Karen Simonyan (2014) Very Deep Convolutional Networks for Large-Scale Image Recognition (VGG) Section4 Link: https://arxiv.org/abs/1409.1556.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Highlights in Science, Engineering and Technology
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
