Wood surface defect detection technology offers the advantages of being non-destructive, rapid, accurate, and economical. It plays a crucial role in wood grade sorting, defect detection, improving the quality of sawn timber, and accelerating the automation of wood processing. Currently, there are challenges in accurately identifying multi-scale wood defects and insufficient overall detection accuracy in the field of wood defect detection. To address these issues, a new wood defect detection model named DRR-YOLO is proposed in this study. This proposed model combines the DWR module and the DRB module to innovatively form the DRRB module, replacing the bottleneck part of the C2f module in the YOLOv8 backbone, thereby constructing the C2f-DRRB module. This module effectively extracts multi-scale feature information. Additionally, by introducing the LSKA attention mechanism, the DRR-YOLO captures a wider range of global information. The neck structure of the DRR-YOLO is reconstructed using BiFPN, further enhancing the integration of feature information. In a series of ablation and comparative experiments, the DRR-YOLO model demonstrates superior performance, with its mean average precision (mAP) improved by 5.2% compared to the original algorithm. This effectively meets the wood industry's demand for accurate detection of wood defects.