鲁荣乔,范 兴,沈立东,张军锦,邓江林

针对烟包表面微小缺陷检测中存在目标尺寸小、背景复杂、传统算法漏检率高等问题，提出一种融合Transformer全局特征建模能力与多尺度注意力机制的改进YOLO算法。通过引入Transformer模块增强主干网络的长程依赖捕捉能力，设计跨尺度通道-空间注意力模块（CSA）优化小目标特征表达，并结合自适应样本加权策略提升模型对微小缺陷的敏感性。实验表明，该方法在自建烟包缺陷数据集上mAP达到92.3%，较基线YOLOv5提升8.5%。

CT-YOLO；烟包缺陷；Transformer

[1] Wang T,Chen L,Li Q.Multi-scale Gabor filter combination for uniform texture defect detection[J].IEEE Transactions on Indust rial Informatics,2015,11(3):632-640. [2] Zhou Y,Zhang H,Wang R.Comparative study of traditional vision algorithms in complex packaging inspection[J].Journal of Int elligent Manufacturing,2017,28(4):1023-1035. [3] Ren S,He K,Girshick R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[C].Advances in Neural Information Processing Systems,2015: 91-99. [4] Lin T Y,Goyal P,Girshick R,et al.Focal loss for dense object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(2):318-327. [5] He K,Gholami A,Girshick R,et al.Mask R-CNN[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(2): 386-397. [6] Lin T Y,Dollár P,Girshick R,et al.Feature pyramid networks for object detection[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2017: 2117-2125. [7] Liu X,Wu Y,Li W.Limitations of single-stage detectors in PCB micro-defect detection[J]. IEEE Transactions on Components,P ackaging and Manufacturing Technology,2019,9(6): 1124-1133.

点击下载PDF