基于改进YOLOv8模型的木材缺陷检测

打开文本图片集

关键词：木材检测；深度学习；损失函数；条件卷积；特征融合；YOLOv8；缺陷识别 中图分类号：TP391.41 文献标识码：A DOI：10.7525/j.issn.1006-8023.2025.04.010

Abstract：Tosolvetheproblemthatthetargetdetectionalgorithmisprone toleakageandlacksdetectionaccuracy ndetecting wood surface defects，this paper proposes an improved YOLOv8 model（YOLOv8-CBW，C，Band Ware abbreviations for CondSiLU，BiFPNand Wise-IoU）and constructs aself-made dataset containing various wood defects.Byoptimizing theoriginal YOLOv8 algorithmandcombining CondConv（conditional convolution）with SiLU（sigmoidweightedlinearunit）to formtheCondSiLUmodule insteadofthetraditionalconvolutionmodule，theflexibilityoffeature extraction is improved;the bidirectionalfeature pyramid network（BiFPN）is introduced toenhancethe multi-scale feature fusioncapability；andthe Wise-IoU（weighted intersection over union）loss functionreplaces the CIoU（complete intersectionoverunion）to improvetheadaptabilityand generalizationperformanceof the model tolow-qualitysamples. The experimental results show that the improved YOLOv8-CBW model improves the mAP5O（mean average precision at IoU threshold O.5O）and mAP50-95（mean average precision over IoU thresholds from 0.50 to 0.95）by 3.7% and （204号 3.9% ，respectively，compared with the YOLOv8 model，and it shows higher precision and stability in complex wood defectdetectiontasks.Theresearch in this paper provides new ideasfor wood defectdetection tasksand has good practical application prospects.

Keywords：Wood detection；deep learning；loss function；conditionalconvolution；feature fusion；YOLOv8；defect identification

0 引言

木材作为一种重要的可再生资源，具有重要的生态价值1和经济价值。（剩余14702字）