Lightweight embedded detection system for historic wooden architecture using an improved YOLOv8 model

Historic wooden architectures are an important component of architectural heritage, but their condition assessment still relies heavily on manual inspection, which is time-consuming and subject to human variability. However, existing methods often struggle to balance detection accuracy and computational efficiency, limiting their applicability in resource-constrained field environments. In this study, a lightweight embedded defect detection system for historic wooden architectures is proposed based on an improved YOLOv8 model. First, a dedicated dataset is constructed, containing multiple defect categories, including scab, crack, micro-organisms, insect damage, and aging, collected under diverse field conditions. Second, a lightweight shared convolutional detection (LSCD) head is introduced to replace the conventional detection head in YOLOv8, reducing parameter redundancy and computational complexity. Third, an embedded detection platform is developed to support image, video, and real-time inspection. The system achieves a real-time inference speed of 293.86 FPS on the embedded device. Experimental results demonstrate that the proposed method maintains stable and accurate detection under challenging conditions such as complex textures and uneven illumination. Field validation further confirms the robustness and practical applicability of the proposed system. These results indicate that the proposed lightweight detection framework provides an efficient and reliable solution for automated heritage timber inspection.