Recommendation for protection with deep learning based method before conservation and repair applications

In order to transfer cultural heritages to future generations, effective and sustainable protection-repair practices are required in the degradation process. First of all, the types of deterioration seen in cultural heritages should be analyzed and appropriate protection-repair practices should be planned. The accuracy and effectiveness of these protection studies for cultural heritages are of great importance. Any mishandling of deterioration can accelerate existing deterioration and cause irreversible problems. For these reasons, it is necessary to develop alternative audit methods in order to reduce the share of human error and to adopt a more universal approach. The Deep Learning based Mask-RCNN algorithm used in the thesis offers an effective solution. For this algorithm, which basically tries to imitate human skills and extract automatic features, a data set was created by collecting images from cultural heritages built with Sille building stone in Konya. The model was trained by labeling appropriate protection suggestions for the deteriorations in these images. The model has been tested on images obtained from Şeyh Osman Rumi Tomb and Hoca Ahmet Fakîh Mosque. When the results were examined, it was seen that the proposed model could provide important foundations for the studies carried out before the protection-repair applications. Thus, with this study, a step was taken to improve the integration of deep learning, one of the subjects of artificial intelligence, into the restoration field.