面向焊缝表面缺陷的双向星聚融合检测网络

doi:10.3778/j.issn.1002-8331.2501-0226

摘要/Abstract

摘要： 针对焊缝表面缺陷检测中的目标存在空间聚集分布、背景纹理复杂、特征显著性不足等挑战，以及传统模型在低信噪比环境下特征表征能力弱、计算冗余等问题，基于YOLOv8n设计星聚双向融合网络（LSBi-YOLO）。构建星聚融合网络（SANet），利用三阶异构卷积融合策略保留关键特征；开发方向增强型空间金字塔池化模块（OS-SPPF），通过线性纹理特征权重优化提升多类缺陷辨识度；进而设计自适应双向多尺度融合网络（ABiMFN），建立跨层级双向交互机制以强化缺陷特征表达并抑制背景噪声；提出融合双重聚焦机制的（PIoU）损失函数，通过抑制冗余检测框有效提升模型收敛速度。实验结果表明，LSBi-YOLO模型在焊缝缺陷检测任务中综合性能显著提升，相较于基线模型，其测试集准确率、召回率和mAP@0.5分别提高3.6、4.6和3.2个百分点，模型参数量减少0.39×106，检测速度达到108.3?FPS；在嵌入式设备上的检测速度为59.2?FPS，平均精度为86.3%，结果证明整体性能优于当前主流焊缝缺陷检测算法。

关键词: 缺陷检测, YOLOv8n, 星型操作, 注意力机制, 加权特征金字塔, 损失函数

Abstract: To address the challenges in weld seam surface defect detection, including spatially aggregated target distributions, complex background textures, and insufficient feature saliency, as well as the weak feature representation and computational redundancy of traditional models in low signal-to-noise ratio environments, this paper proposes a lightweight bidirectional fusion network (LSBi-YOLO) based on the YOLOv8n architecture. Firstly, the star convergence fusion network (SANet) is constructed, and its key features are preserved through a third-order heterogeneous convolution fusion strategy. Secondly, an orientation-enhanced spatial pyramid pooling module (OS-SPPF) is developed to optimize the weight distribution of linear texture features, thereby improving the recognition accuracy of multiple types of defects. Furthermore, an adaptive bidirectional multi-scale fusion network (ABiMFN) is designed to establish a cross-level bidirectional interaction mechanism, enhancing defect feature expression while suppressing background noise. Finally, a dual-focusing PIoU loss function is introduced to suppress redundant detection boxes, thereby accelerating model convergence. The experimental results demonstrate that the LSBi-YOLO model achieves a significant enhancement in comprehensive performance for weld defect detection. Compared with the baseline model, the test set accuracy, recall rate, and mAP@0.5 are improved by 3.6, 4.6, and 3.2 percentage points, respectively. Additionally, the model reduces the number of parameters by 0.39×106 while achieving a detection speed of 108.3?FPS. The embedded device achieves a detection speed of 59.2?FPS with an average accuracy of 86.3%. These results demonstrate that the overall performance surpasses that of the current mainstream weld defect detection algorithms.

Key words: defect detection, YOLOv8n, star operation, attention mechanism, weighted feature pyramid, loss function

徐成, 单文桃, 张陈, 陈泽明, 杨汶睿, 韩振华. 面向焊缝表面缺陷的双向星聚融合检测网络[J]. 计算机工程与应用, 2026, 62(8): 270-284.

XU Cheng, SHAN Wentao, ZHANG Chen, CHEN Zeming, YANG Wenrui, HAN Zhenhua. Bidirectional Star Fusion Detection Network for Weld Surface Defects[J]. Computer Engineering and Applications, 2026, 62(8): 270-284.

参考文献

[1] 胡丹，高向东，张南峰，等.焊缝缺陷检测现状与展望综述[J].机电工程，2020，37（7）：736-742.
HU D，GAO X D，ZHANG N F，et al.Review of status and prospect of weld defect detection[J].Mechanical & Electrical Engineering Magazine，2020，37（7）：736-742.
[2] 秦飞红，袁懋诞，祝端阳，等.基于多层特征融合目标检测网络的超声全聚焦钢板焊缝缺陷自动检测研究[J].机械工程学报，2025，61（4）：55-66.
QIN F H，YUAN M D，ZHU D Y，et al.Automatic weld defect detection in steel plates based on ultrasonic total focusing method and multi-layer feature fusion target detection network[J].Journal of Mechanical Engineering，2025，61（4）：55-66.
[3] GIRSHICK R，DONAHUE J，DARRELL T，et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2014：580-587.
[4] GIRSHICK R.Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision.Piscataway：IEEE，2016：1440-1448.
[5] REN S Q，HE K M，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149.
[6] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot MultiBox detector[C]//Proceedings of the 14th European Conference on Computer Vision.Cham：Springer，2016：21-37.
[7] BOCHKOVSKIY A，WANG C Y，LIAO H M.YOLOv4：optimal speed and accuracy of object detection[EB/OL].[2024-12-02].https：//arxiv.org/abs/2004.10934.
[8] 张惠莉，代晨龙，任景龙，等.基于YOLO v8-GSGF模型的葡萄病害识别方法研究[J].农业机械学报，2024，55（11）：75-83.
ZHANG H L，DAI C L，REN J L，et al.Grape disease identification method based on YOLO v8-GSGF model[J].Transactions of the Chinese Society for Agricultural Machinery，2024，55（11）：75-83.
[9] 毛清华，郭文瑾，苏毅楠，等.改进RT-DETR的煤矿刮板输送机链条故障智能识别方法[J].煤炭科学技术，2025，53（9）：469-479.
MAO Q H，GUO W J，SU Y N，et al.Improvement of the RT-DETR-Based intelligent identification method for chain faults in coal mine scraper conveyors[J].Coal Science and Technology，2025，53（9）：469-479.
[10] SAFALDIN M，ZAGHDEN N，MEJDOUB M.An improved YOLOv8 to detect moving objects[J].IEEE Access，2024，12：59782-59806.
[11] 梁礼明，龙鹏威，李俞霖.改进轻量高效FMG-YOLOv8s的钢材表面缺陷检测算法[J].计算机工程与应用，2025，61（3）：84-93.
LIANG L M，LONG P W，LI Y L.Improved lightweight and efficient FMG-YOLOv8s algorithm for steel surface defect detection[J].Computer Engineering and Applications，2025，61（3）：84-93.
[12] MA X，DAI X Y，BAI Y，et al.Rewrite the stars[C]//Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2024：5694-5703.
[13] XU W，WAN Y.ELA：efficient local attention for deep convolutional neural networks[EB/OL].[2024?12?05].https：//arxiv.org/abs/2403.01123.
[14] WANG H，ZHANG F，WANG L.Fruit classification model based on improved Darknet53 convolutional neural network[C]//Proceedings of the 2020 International Conference on Intelligent Transportation，Big Data & Smart City.Piscataway：IEEE，2020：881-884.
[15] HOU Q B，ZHANG L，CHENG M M，et al.Strip pooling：rethinking spatial pooling for scene parsing[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2020：4002-4011.
[16] WANG K X，LIEW J H，ZOU Y T，et al.PANet：few-shot image semantic segmentation with prototype alignment[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision.Piscataway：IEEE，2019：9196-9205.
[17] GHIASI G，LIN T Y，LE Q V.NAS-FPN：learning scalable feature pyramid architecture for object detection[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2020：7029-7038.
[18] CHEN J，MAI H S，LUO L B，et al.Effective feature fusion network in BIFPN for small object detection[C]//Proceedings of the 2021 IEEE International Conference on Image Processing.Piscataway：IEEE，2021：699-703.
[19] ZHENG Z H，WANG P，REN D W，et al.Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J].IEEE Transactions on Cybernetics，2022，52（8）：8574-8586.
[20] YU J H，JIANG Y N，WANG Z Y，et al.UnitBox：an advanced object detection network[C]//Proceedings of the 24th ACM International Conference on Multimedia.New York：ACM，2016：516-520.
[21] REZATOFIGHI H，TSOI N，GWAK J，et al.Generalized intersection over union：a metric and a loss for bounding box regression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2020：658-666.
[22] ZHANG Y F，REN W Q，ZHANG Z，et al.Focal and efficient IOU loss for accurate bounding box regression[J].Neurocomputing，2022，506：146-157.
[23] ZHENG Z H，WANG P，LIU W，et al.Distance-IoU loss：faster and better learning for bounding box regression[C]//Proceedings of the 34th AAAI Conference on Artificial Intelligence.Palo Alto：AAAI，2020：12993-13000.
[24] LIU C，WANG K G，LI Q，et al.Powerful-IoU：more straightforward and faster bounding box regression loss with a nonmonotonic focusing mechanism[J].Neural Networks，2024，170：276-284.
[25] WEN X，SHAN J，HE Y，et al.Steel surface defect recognition：a survey[J].Coatings，2022，13（1）：17.
[26] 苏志威，黄子涵，邱发生，等.基于改进YOLOv8的航空铝合金焊缝缺陷检测方法[J].航空动力学报，2024，39（6）：116-124.
SU Z W，HUANG Z H，QIU F S，et al.Weld defect detection of aviation aluminum alloy based on improved YOLOv8[J].Journal of Aerospace Power，2024，39（6）：116-124.
[27] 周翌晨，虞旦，李佳成，等.基于改进YOLOv5的焊缝识别算法研究[J].制造技术与机床，2024（7）：19-25.
ZHOU Y C，YU D，LI J C，et al.Research on weld identification algorithm based on improved YOLOv5[J].Manufacturing Technology & Machine Tool，2024（7）：19-25.
[28] 岳剑峰，李伟明，宁黎华，等.基于YOLO-DEFW的焊缝缺陷实时检测算法研究[J].中国激光，2025，52（8）：56-68.
YUE J F，LI W M，NING L H，et al.Real-time weld defect detection algorithm based on YOLO-DEFW[J].Chinese Journal of Lasers，2025，52（8）：56-68.
[29] CHEN D F，ZHANG L C.SL-YOLO：a stronger and lighter drone target detection model[EB/OL].[2024?12?18].https：//arxiv.org/abs/2411.11477.
[30] HE K M，GKIOXARI G，DOLLáR P，et al.Mask R-CNN[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision.Piscataway：IEEE，2017：2980-2988.
[31] ZHAO Y A，LV W Y，XU S L，et al.DETRs beat YOLOs on real-time object detection[C]//Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2024：16965-16974.
[32] ZHANG H Y，WANG Y，DAYOUB F，et al.VarifocalNet：an IoU-aware dense object detector[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway：IEEE，2021：8510-8519.
[33] 李明，何志奇，党青霞，等.面向户外导盲场景的道路目标检测算法[J].计算机工程与应用，2025，61（9）：242-254.
LI M，HE Z Q，DANG Q X，et al.Road object detection algorithm for outdoor blind navigation scenarios[J].Computer Engineering and Applications，2025，61（9）：242-254.