AN APPROACH WITH DEEP CONVOLUTIONAL NEURAL NETWORKS FOR ACCURATE ARCHITECTURAL STYLE CLASSIFICATION

Unlike other studies, this study aims to examine the effect of data diversity and quantity on the performance of a model rather than solely focusing on classifying architectural styles. It also seeks to emphasize the importance of selecting the appropriate number of classes in architectural style classification and to explore how determining the optimal number of classes affects the model's success. The first dataset comprises the original images (4776 images), the second dataset (10091 images) includes additional data and the third dataset (9552 images) is obtained through data augmentation. Using these three datasets, models were developed by reducing the number of classes with convolutional neural networks (CNNs), which utilize convolution layers to identify local features in the images and classify them by summarizing these features. The results of the developed models show that the method of adding data and increasing data decreased the success of architectural style classification, whereas reducing the number of styles increased it. This study may lay the groundwork for the development of deep learning models for future architectural style classification. The findings of this study can significantly impact various application areas, such as analyzing architectural styles, identifying historical periods and integrating them into architectural education.