Real-time detection of cracks: improved YOLOv8 network applied to wooden structure of residential historic buildings

Residential historical buildings carry rich historical information and cultural connotations, as well as being an important part of China's architectural engineering. However, over time, cracks have developed in the wooden structures of these buildings, posing a threat to the structural integrity and long-term preservation of the buildings. Timely detection and repair of these cracks is therefore critical. In this study, we propose and apply an improved YOLOv8-n model for real-time detection of cracks in wooden members of residential historic buildings. First, we collected a large image dataset of cracks in wooden structure of residential historic buildings and labeled the cracks. Then, we improved the YOLOv8-n model by introducing the coordinate attention mechanism and the Ghost convolution module so that the model can pay more attention to the details and features of the crack region. Finally, we completed the training and testing of the YOLOv8-n model before and after the improvement on the constructed wood member crack dataset. The experimental results show that the improved YOLOv8-n model achieves significant performance improvement in the detection of cracks in wood members of residential historic buildings compared with the original YOLOv8-n model, with an F1 score of 94.7%, an average accuracy of 96.2%, and a detection speed of 89.3fps. In conclusion, compared with the traditional contact detection method, our proposed improved YOLOv8-n inspection model is a contactless nondestructive inspection method. The method provides fast and accurate detection of wood structure cracks and accurate crack localization information, providing strong support for the restoration and preservation of crack-damaged wood structures in residential historic buildings.