Detecting Defects of Wooden Boards by Improved YOLOv4-tiny Algorithm

Wooden boards are widely used as raw materials for furniture. Detecting defects on the wooden is significant to improve the quality of products. In this paper, we propose an improved YOLOv4-tiny target detection algorithm with feature extraction and utilization tailored for wood defect recognition. The proposed algorithm replace the residual network of the original algorithm with the Res2Net network to enhance the feature extraction ability, and increase the different scales to improve the prediction accuracy. A testbed is deployed to collect the images of wooden boards, and the performance of the proposed algorithm is evaluated on the testbed. Compared with the original YOLOv4-tiny algorithm, the mean average precision for detecting the three types of defects deadknot, insectpest and crack was 80.06% which was an increase of 5%. Compared with other methods of wood defect detection, the proposed algorithm is more accurate and faster.