A deep learning integrated mobile application for historic landmark recognition: A case study of Istanbul

Abstract

<p><span style="color: rgb(51, 51, 51); font-family: &quot;Helvetica Neue&quot;, Helvetica, Arial, sans-serif; font-size: 12.6px;">Recent developments in mobile device technology and artificial intelligent systems took the attention of many researchers. Historical sites and landmarks are the indispensable heritage of cities. Historic landmark recognition, including detailed attribute information, can connect people directly with the history of the cities, although they may not be familiar with the impressive historical monument. This can be achieved by integrating mobile and deep learning technologies. Therefore, we focused on establishing a deep learning (DL) based mobile historic landmark recognition system in this study. The VGG (16, 19), ResNet (50, 101, 152), DenseNet (121, 169, 201) DL architectures were trained by end-to-end learning techniques for the recognition of ten historic landmarks from the metropolitan city of Istanbul, Turkey. The dataset was prepared by collecting images of ten historical buildings from the image hosting services. The developed prototype automatically and instantly recognizes these historic landmarks from scene images and immediately provides related historic information as well as route planning. The experimental results indicate that DenseNet-169 architecture is very effective for our dataset with 96.3% accuracy. This study has shown that deep learning offers a promising alternative means of recognizing historic landmarks.</span><br></p>