Research on the application of loss function in wood surface defect recognition algorithm

China is a big country in the production, manufacturing and export of wood and wood products, and the surface defects of wood will reduce the external quality and internal strength of wood, and then affect the processing and production process of wood. Based on the self-made wood surface defect data set, this paper conducts training, testing and comparison for the current mainstream detection algorithm models (SSD-300, Faster-Rcnn, YOLOv3, YOLOv4-csp and YOLOv5s). Then, on the basis of YOLOv5s model, 10 kinds of popular loss functions are tested and analyzed comprehensively. The purpose of this paper is to summarize the existing loss functions and expand the application of deep learning model in the field of wood defect detection, in order to provide certain guidance and reference for the automatic detection of wood surface defects.