Intelligent Detection of Deterioration in Cultural Stone Heritage

Abstract

Vision-based periodic examination of the deterioration of stone monuments over time is labour and time intensive. Especially, in cases involving large-scale immovable cultural heritage, the workforce is considerably increased, along with the possibility of occurrence of errors. Any misdiagnoses in the deterioration may cause irreversible structural problems in monuments, and thus, it is necessary to develop alternative examination methods. Computer-vision methods represent an effective solution to eliminate both human errors and difficulties in the field. Therefore, this study aims to adopt the Mask R-CNN algorithm, which is a computer-vision method, to detect and map the deteriorations observed in the Gumus, ler archaeological site and monastery (cracks, discontinuities, contour scaling, missing parts, biological colonization, presence of higher plants, de-posits, efflorescence, and loss of fresco). First, 1740 images were collected from the site, and the model was trained by labelling the distortions in these images according to their types. Later, the model was tested on four outdoor and two indoor views. The developed model achieved an average precision ranging between 91.591% and 100%, and the mean average precision was 98.186%. These results demonstrated that the proposed algorithm can enable mapping to promptly and automatically detect the deterioration in large monuments.