RBINet: A Convolutional Neural Network for Automatic Building Identification

Convolutional neural networks (CNNs) have been used to analyze visual data in several applications related to computer vision, including image classification and segmentation. The aim of this paper is to use a trained CNN model to accurately recognize various building from images. However, the exploration of this task has been hampered by the lack of well-organized, sizable architectural picture datasets with a particular focus on building identification. To address this, we created a dataset of building pictures with related labels to train and test our CNN model for building identification. The name of the dataset is the RUET-building image (RBI) dataset as the photographed buildings are located at the RUET (a prestigious engineering University) campus. There are 7075 photos in this collection, which were gathered from 15 distinct buildings. In order to find the ideal model for the RBI dataset, this work also contains an ablation analysis of our proposed CNN named RBINet (RUET-building identification network). In terms of building recognition accuracy, RBINet (with four convolutional layers and three hidden layers) outperforms leading models and achieves 98.88% accuracy. The authors also report that the EfficientNetb0 and DenseNet121 models give the second (98.44%) and third (98.22%) highest classification accuracy, respectively, for the RBI dataset.