【摘要】 针对YOLOv8s模型在焊缝缺陷检测中误检、漏检以及特征提取、特征融合能力不足的问题，文章提出了YOLOv8s改进算法。该算法通过将原网络模型中的部分C2f模块升级为C2f-Triplet模块，增强了特征提取和融合能力；引入CARAFE上采样算子，提升了大范围特征信息的聚合；并将CIoU损失函数改进为Powerful-IoU,加快了收敛并提高了回归精度。实验结果表明在Weld Defect.v1i.yolov8数据集上，改进YOLOv8s算法较原算法精度提高1.6百分点，平均精度均值提高2.6百分点，可以更加准确地检测出焊缝缺陷的类别和位置。更多还原

【Abstract】 Aiming at the problems of misdetection, omission and insufficient feature extraction and feature fusion ability of YOLOv8s model in weld defect detection, this paper proposes an improved algorithm for YOLOv8s.The algorithm enhances the feature extraction and fusion ability by upgrading part of the C2f module in the original network model to the C2f-Triplet module; introduces the CARAFE up-sampling operator to improve the aggregation of a large range of feature information; and improves the CIoU loss function to the Powerful-IoU,which accelerates the convergence and improves the regression accuracy.The experimental results show that on the Weld Defect.v1i.yolov8 dataset, the improved YOLOv8s algorithm improves the accuracy of the original algorithm by 1.6 percentage points, and the mean of the average accuracy improves by 2.6 percentage points, which can more accurately detect the category and location of weld defects.更多还原