Research on intelligent monitoring technology for roof damage of traditional Chinese residential buildings based on improved YOLOv8: taking ancient villages in southern Fujian as an example

Abstract

Abstract In the process of preserving historical buildings in southern Fujian, China, it is crucial to provide timely and accurate statistical data to classify the damage of traditional buildings. In this study, a method based on the improved YOLOv8 neural network is proposed to select aerial photographs of six villages in Xiamen and Quanzhou cities in Fujian Province as the dataset, which contains a total of 3124 photographs. Based on the high-resolution orthophotographs obtained from UAV tilt photography, the YOLOv8 model was used to make predictions. The main task in the first stage is to select the buildings with historical value in the area, and the model's mAP (Mean Accuracy Rate) can reach 97.2% in the first stage task. The second stage uses the YOLOv8 model to segment the images selected in the first stage, detecting possible defects on the roofs, including collapses, missing tiles, unsuitable architectural additions, and vegetation encroachment. In the second stage of the segmentation task, the mAP reaches 89.4%, which is a 1.5% improvement in mAP50 (mean accuracy) compared to the original YOLOv8 model, and the number of parameters and GFLOPs are reduced by 22% and 15%, respectively. This method can effectively improve the disease detection efficiency of historical built heritage in southern Fujian under complex terrain and ground conditions.

Keywords

• YOLOv8
• UAV
• Historic buildings
• Deep learning
• Object detection