ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Journal Article A Deep Learning Framework for Cultural Heritage Damage Detection for Preservation; Based on the Case of Heunginjimun and Yeongnamnu in South Korea
Cited 0 time in scopus Share share facebook twitter linkedin kakaostory
Authors
Sang-Yun Lee, Daekyeom Lee
Issue Date
2023-07
Citation
ICACT Transactions on Advanced Communications Technology, v.12, no.4, pp.1507-1513
ISSN
2288-0003
Publisher
Global IT Research Institute
Language
English
Type
Journal Article
DOI
https://dx.doi.org/10.23919/ICACT60172.2024.10471765
Abstract
In general, there are many restrictions on investigations for safety inspection due to the uniqueness of cultural heritages. Methods such as visual inspection and non-destructive inspection, which are mainly used as inspection methods, are regularly carried out, but there are limitations on time and cost. This is insufficient to identify and respond quickly when an abnormal symptom appears in cultural heritage. As a basic study of system development for rapid abnormal detection of architectural, cultural properties through Deep Learning, this paper organized a Deep Learning framework for detecting tilt in buildings for the roof of Heunginjimun Gate (Korea Treasure No.1) and Yeongnamnu Pavilion (Korea Treasure No. 147). A framework was developed using a Convolutional Neural Network (CNN). As a result of an application, EfficientnetBO and EfficientnetB2 models showed excellent accuracy in detecting the tilt of the roof of Heunginjimun with an average accuracy of 99.66% and 99.69%, respectively. In addition, EfficientnetBO, EfficientnetB2, and Shufflenet_ v2 models showed excellent accuracy in detecting tilt of the roof of Yeongnamnu with 98.81 %, 99.80%, and 98.48% accuracy. Additionally, the Grad-CAM experiment was conducted as a basis for whether the model made the proper judgment to confirm the criteria for determining abnormal detection according to the results of each model. These findings quickly detect abnormalities occurring in cultural heritages from the perspective of cultural heritage management and preservation, enabling rapid response, and are valuable for research on artificial intelligence technology related to cultural heritages.
KSP Keywords
Artificial intelligence technology, Convolution neural network(CNN), Cultural heritage management, Deep learning framework, Non-Destructive Inspection, Rapid Response, Safety inspection, South Korea, abnormal detection, damage detection, deep learning(DL)