The Study of Deep Learning for Detecting Architectural Heritage Defects

Abstract

Heritage buildings are valuable historic structures that play an increasingly important role in reflecting history, promoting cultural inheritance, and displaying values, which makes their conservation attract increasing attention from the international community. Nevertheless, the situation of heritage building conservation in China looks grim.The study introduces the relevant research objects and research methods involved in the topic in detail and explains the related concepts such as architectural pathology and the types of historical building diseases. The research examined the usage of deep convolutional neural networks (DCNNs) for the classification, segmentation, and detection of the images of surface defects in heritage buildings. A survey was conducted on the building surface defects in Gulang Island (a UNESCO World Cultural Heritage Site), which were subsequently classified into six categories according to relevant standards. A Swin Transformer- and YOLOv5-based model was built for the automated detection of surface defects. Experimental results suggested that the proposed model was 99.2% accurate at classifying plant penetration and achieved a mean intersection-over-union (mIoU) of over 92% in relation to moss, cracking, alkalization, staining, and deterioration, outperforming CNN-based semantic segmentation networks such as FCN, PSPNet, and DeepLabv3plus. The Swin Transformer-based approach to the segmentation of building surface defect images achieved the highest accuracy regardless of the evaluation metric (with a mIoU of 90.96% and a mAcc of 95.78%), when contrasted to mainstream DCNNs such as SegFormer, PSPNet, and DANet.The study also examines the application of the YOLOv5- and Swin Transformer-based deep learning model in recognizing images of surface defects in heritage buildings. An improved YOLOv3 algorithm based on deep learning is proposed to address the problems of small target miss detection, high background complexity and many false alarm interference factors encountered in remote sensing image architectural heritage disease recognition. Based on the first-stage representative algorithm YOLOv3, the detection capability and accuracy of the algorithm are enhanced by improving the network structure, modifying the feature map scale and embedding the null convolution module. The experimental data show that the improved YOLOv3 algorithm has improved the average detection accuracy and recall rate compared with the original algorithm on the basis of satisfying the real-time performance, and thus the proposed method has effectively improved the detection effect of remote sensing architectural heritage diseases.

-