Fine-Tuned Convolutional Neural Networks for Bangladeshi Vehicle Classification

Classification of vehicles plays an important role in the intelligent transport system. In this paper, we propose a framework for the classification of Bangladeshi vehicle images based on fine-tuned convolutional neural networks (CNNs). Using the proposed framework and an unified experimental setting, we fine-tuned fifteen popular CNN architectures, namely, AlexNet, Inception-V3, VGG-II, VGG-13, VGG-16, VGG-19, ResNet18, ResNet-34, ResNet-50, ResNet-101, ResNet-152, DenseNet121, DenseNet-161, DenseNet-169 and DenseNet-201 pretrained on ImageNet dataset on a public Bangladeshi vehicle image dataset. We conduct a systematic and comprehensive analysis on the performance of the fine-tuned CNNs which leads to new insights. Our experimental results show that ResNet-152 and DenseNet-201 fine-tuned with our proposed strategy provide excellent performance for the classification of Bangladeshi vehicle images. Our implementation is available at https://github.con MinhajurRFahad/bd-vehicle-cnn-benchmark.