Automatic detection of wood cracks based on improved YOLOv8 algorithm

Affected by climate and human factors, wood cracks represent the most prevalent form of damage in wooden heritage structures. Due to the time-consuming and labor-intensive nature of traditional detection methods, this study employs the YOLOv8 algorithm. To effectively reduce network complexity without compromising recognition accuracy during YOLOv8 model training, a lightweight enhancement of the YOLOv8 network, inspired by the slim-neck concept, is proposed to improve target detection efficiency. Additionally, a dataset consisting of 406 images of cracks is constructed, and four models from the YOLO series are employed for discussion. The final experiment concludes that the performance of the improved YOLOv8 is enhanced, and the model complexity is also reduced. The precision of the improved model is up to 95.13% and the mAP50 of mean average precision (mAP) is 94.5%.