Distress detection and classification of archaeological monuments through deep learning: A case study of chellah, a Moroccan monument

Abstract

Chellah, a Moroccan historical monument, possesses profound cultural, economic, and archaeological significance. This site represents a seamless blend of Islamic heritage and modern Western influences, not only preserving the remnants of ancient civilizations like the Phoenician, Carthaginian, and Roman but also vividly portraying diverse lifestyles and numerous legends within its walls. This remarkable monument withstands the relentless forces of nature, enduring both physical wear and chemical degradation, which results in the deterioration of its structural integrity and poses a threat to its safety. Therefore, regularly assessing this edifice is crucial to guarantee the preservation and upkeep of this historical monument, preventing its deterioration. Employing visual inspections conducted manually to detect and classify the different distress in historical monuments demonstrates itself as a labor-intensive and time-consuming endeavor. In response to these limitations, the current research presents a novel damage detection method for the automated identification of deterioration in Chellah, with the objective of accelerating the inquiry process and optimizing the effectiveness of distress identification. This study introduces a pioneering approach for automated damage detection in historical monuments, specifically targeting the Chellah site in Morocco. Leveraging the YOLOv5 deep learning model, this research achieves exceptional precision (97 %) and F1 score (92 %), outperforming state-of-the-art models like YOLOv7 and YOLOv8. Unlike traditional methods reliant on costly equipment or labor-intensive manual inspections, this method addresses challenges such as the detection of small or overlapping damages and the efficient use of a relatively small dataset. The novelty of this work lies in tailoring advanced object detection technologies to the complex, irregular surfaces of the Chellah monument, demonstrating superior real-time performance and low computational cost. This contribution provides a robust, scalable solution for preserving cultural heritage and sets a benchmark for future applications in heritage conservation and real-time monitoring.