Road-type classification is increasingly becoming important to be embedded in interactive maps to provide additional useful information for users. The ubiquity of smartphones supported with high definition cameras offers a rich source of information that can be utilised by machine learning techniques. In this paper, we propose a novel Convolutional Neural Network (CNN)-based approach to classify road types using a collection of publicly available images. To overcome the challenge of having huge dataset to train and test CNNs, our approach employs fine-tuning. We conducted an experiment where the VGG-16, VGG-S and GoogLeNet networks were constructed and fine-tuned with the dataset gathered. Our approach achieved an accuracy of 99% in VGG-16 and 100% in VGG-S, while using the GoogLeNet model produced results up to 98%.
Paper
Full text
Road-Type Classification through Deep Learning Networks Fine-Tuning
Semantic Scholar · Computer Science · 2020
Abstract
Road-type classification is increasingly becoming important to be embedded in interactive maps to provide additional useful information for users. The ubiquity of smartphones supported with high definition cameras offers a rich source of information that can be utilised by machine learning techniques. In this paper, we propose a novel Convolutional Neural Network (CNN)-based approach to classify road types using a collection of publicly available images. To overcome the challenge of having huge dataset to train and test CNNs, our approach employs fine-tuning. We conducted an experiment where the VGG-16, VGG-S and GoogLeNet networks were constructed and fine-tuned with the dataset gathered. Our approach achieved an accuracy of 99% in VGG-16 and 100% in VGG-S, while using the GoogLeNet model produced results up to 98%.