Wood surface defect detection poses challenges due to the diverse range of defects, making accurate localization and identification difficult. In this study, we introduce an enhanced approach for detecting flaws on wood surfaces by leveraging an augmented version of the YOLOv8s algorithm. To improve the focus on problematic target qualities, we initially constructed a HAM (hybrid attention module) structure within the Backbone. This structure incorporates spatial and channel attention techniques, enhancing the ability to identify defects. Additionally, we enhance the feature fusion capabilities by augmenting the expansion convolution module, reducing information loss during the connection with the Neck network. This augmentation improves the target receptive field, ensuring critical information preservation for effective diagnosis of wood surface defects. Furthermore, we introduce ghost convolution to enhance feature expression while minimizing the number of parameters. This approach optimizes the model's overall performance. Through extensive testing, our proposed GH-YOLOv8s model demonstrates accurate detection of five distinct types of wood surface defects, including defect types such as Live Knot, Dead Knot, Resin, Knot with crack, and Crack, achieving a mean average precision (mAP) of 98.4%. This performance surpasses the original model by 2.0% and maintains a high FPS (Frames Per Second) rate of 163.9, this means it can achieve efficient object detection in real-time scenarios. Moreover, our approach outperforms commonly used target detection methods, establishing its superiority in wood surface defect detection.