YOLOv8-OCHD: A Lightweight Wood Surface Defect Detection Method Based on Improved YOLOv8

To overcome the drawbacks of manual defect detection and the challenges of traditional visual inspection algorithms, such as high missed detection rates, slow detection speeds, and difficulties in deployment on embedded devices for detecting subtle wood surface defects, this study aims to improve the utilization efficiency of wood in enterprises. A lightweight wood surface defect detection method, YOLOv8-OCHD, based on deep learning models, is proposed. This method demonstrates significant performance improvement in handling common multi-scale and small target defects on wood surfaces. Firstly, to enhance the ability to capture multi-dimensional features of wood surface defects and reduce information loss, a fully dynamic convolution (ODConv) is introduced. Secondly, a C2f_RVB module is designed, which uses the RepViTBlock technique to optimize feature representation and effectively reduce the number of model parameters. By enhancing the expression of deep features, the C2f_RVB module significantly improves the accuracy of detecting small target defects on wood surfaces. Next, the Haar Wavelet Downsampling module (HWD) is employed to expand the receptive field, reduce model complexity, and improve both training and inference speeds. Finally, at the output stage, a DyHead (Dynamic Head) attention mechanism detection head is introduced to enhance the detection capability of the algorithm by integrating scale perception, spatial perception, and task perception, significantly improving the feature expression ability at the output. Experimental results show that compared to the YOLOv8n baseline model, the proposed method improves detection accuracy for eight defect types in different tree species, with the mean average precision (mAP) increased by 5.9%. Compared with the YOLOv8n algorithm, the model size is significantly reduced, with the parameter count at only 2.9M and the computational load at 7.0G, representing reductions of 21.6% and 15.6%, respectively. This effectively reduces the difficulty and cost of deployment on mobile terminals while significantly improving algorithm accuracy, meeting real-time requirements, and providing a more efficient and feasible technical solution for relevant applications. These findings suggest that the proposed model shows superior performance in accurately detecting wood surface defects on mobile terminals.