The United Nations aims to preserve, evaluate, and conserve cultural heritage (CH) structures as part of sustainable development. The design life expectancy of many CH structures is slowly approaching its end. It is thus imperative to conduct frequent visual inspections of CH structures following conservation guidelines to ensure their structural integrity. This study implements a custom defect detection, and localization supervised deep learning model based on the you only look once (YOLO) v5 real-time object detection algorithm by implementing a case study of the Dadi-Poti tombs in Hauz Khas Village, New Delhi. The custom YOLOv5 model is trained to automatically detect four defects, namely, discoloration, exposed bricks, cracks, and spalling, and tested on a dataset comprising 10291 images. The validity and performance of the custom YOLOv5 model are compared with a ResNet 101 architecture-based faster region-based convolutional neural network (R-CNN), and conventional manual visual inspection methods are used to convey the significance of the developed artificial intelligence-based model. The maximum average precision (mAP) of the custom YOLOv5 model and faster R-CNN is 93.7% and 85.1%, respectively.

Artificial intelligence-based visual inspection system for structural health monitoring of cultural heritage

Mishra, Mayank;
2022-01-01

Abstract

The United Nations aims to preserve, evaluate, and conserve cultural heritage (CH) structures as part of sustainable development. The design life expectancy of many CH structures is slowly approaching its end. It is thus imperative to conduct frequent visual inspections of CH structures following conservation guidelines to ensure their structural integrity. This study implements a custom defect detection, and localization supervised deep learning model based on the you only look once (YOLO) v5 real-time object detection algorithm by implementing a case study of the Dadi-Poti tombs in Hauz Khas Village, New Delhi. The custom YOLOv5 model is trained to automatically detect four defects, namely, discoloration, exposed bricks, cracks, and spalling, and tested on a dataset comprising 10291 images. The validity and performance of the custom YOLOv5 model are compared with a ResNet 101 architecture-based faster region-based convolutional neural network (R-CNN), and conventional manual visual inspection methods are used to convey the significance of the developed artificial intelligence-based model. The maximum average precision (mAP) of the custom YOLOv5 model and faster R-CNN is 93.7% and 85.1%, respectively.
2022
File in questo prodotto:
File Dimensione Formato  
s13349-022-00643-8.pdf

solo utenti autorizzati

Tipologia: Pdf editoriale
Licenza: DRM non definito
Dimensione 3.98 MB
Formato Adobe PDF
3.98 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11563/193880
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 28
  • ???jsp.display-item.citation.isi??? ND
social impact