"Deep Learning-Based Identification of City Walls in the Morphological Analysis of Khorasan’s Urban Centers During the Middle Islamic Period"

Document Type : Original Article

Authors

1 Ph.D. Student in Archaeology, Department of Archaeology, Faculty of Art and Architecture, University of Mazandaran, Babolsar, Iran

2 Associate Professor, Department of Archaeology, Faculty of Art and Architecture, University of Mazandaran, Babolsar, Iran.

3 ssistant Professor, Department of Archaeology, Faculty of Art and Architecture, University of Mazandaran, Babolsar, Iran

Abstract

In the study of urban spatial structures, the Sharestan wall (urban enclosure) is one of the most significant physical elements of cities during the historical and Islamic periods. This thick wall, especially in the Middle Islamic period, separated the shar (main city area) and its internal components such as residential quarters and bazaars from the Rabad (suburban area), and served as a protective and defensive boundary for the Sharestan. Today, to save time and improve efficiency, automated methods can be used to identify urban elements like city walls. These approaches typically fall within the domain of neural networks and machine learning systems. This study aims to detect city enclosures in the landscapes of historical cities in Khorasan during the Middle Islamic period using a Convolutional Neural Network (CNN) and the YOLOv8 algorithm. Aerial photographs from the 1960s and 1990s (solar Hijri decades of the 1340s and 1370s) were used to build the dataset in order to develop an automated pattern recognition system for identifying city walls in the historical landscapes of Khorasan. The CNN was trained using 80% of the data for training and 20% for validation, over 400 iterations, with a learning rate of 0.01. The results demonstrate the effectiveness of aerial imagery in deep learning for the automatic detection and extraction of destroyed Sharestan walls, achieving an accuracy of 91% with a 9% error rate, and an accuracy of 77% with a 23% error rate for intact walls.

Keywords

Files

History

  • Receive Date: 08 May 2025
  • Accept Date: 08 May 2025

Share

How to cite

Statistics