目的 钢容器作为重要的包装与储运载体,其表面质量直接影响产品质量与使用性能。针对钢表面缺陷形态复杂、方向特征差异明显,以及现有检测模型对不同方向特征利用不足的问题,提出一种基于Mamba-YOLO的钢容器表面缺陷视觉检测方法。方法 在Mamba-YOLO的SS2D模块基础上改进多方向扫描特征融合方式。首先,根据训练集缺陷目标的方向分布构建固定方向先验并引入可学习校正参数。其次,结合自适应方向门控,根据输入特征调整各扫描方向的融合权重,构建Mamba-YOLO-GatePrior模型。采用NEU-DET与GC10-DET两个公开数据集进行训练和测试并与YOLOv8n、YOLO11n和YOLO26n等目标检测模型进行对比。结果 实验结果表明,Mamba-YOLO-GatePrior在NEU-DET和GC10-DET数据集上的mAP50分别达到76.72%和74.98%,较Mamba-YOLO基线分别提高2.05和2.29个百分点。消融实验与对比实验表明,该模型在不同缺陷类型下具有较好的综合检测性能,同时保持较好的推理效率。结论 所提出的钢容器表面缺陷检测方法能够有效提升复杂工业场景下的缺陷识别能力,实现检测性能与推理效率之间的较好平衡,可为钢容器表面缺陷智能检测及包装容器质量控制提供技术参考。
Abstract
Steel containers are important carriers for packaging, storage, and transportation, and their surface quality directly affects product quality and service performance. To address the complex morphology and distinct directional characteristics of steel surface defects, as well as the insufficient utilization of directional features by existing detection models, the work aims to propose a Mamba-YOLO-based visual detection method for steel container surface defects. The multi-directional feature fusion in the SS2D module of Mamba-YOLO was improved. Firstly, a fixed directional prior was constructed from the directional distribution of defect targets in the training set and learnable correction parameters were introduced. Secondly, combined with the adaptive directional gating mechanism, the fusion weights of different scanning directions were adjusted according to the input features, thereby forming the Mamba-YOLO-GatePrior model. The model was trained and tested on the NEU-DET and GC10-DET datasets and compared with target detection models such as YOLOv8n, YOLO11n, and YOLO26n. Experimental results showed that Mamba-YOLO-GatePrior achieved mAP50 values of 76.72% and 74.98% on NEU-DET and GC10-DET, respectively, representing improvements of 2.05 and 2.29 percentage points over the Mamba-YOLO baseline. Ablation and comparative experiments further demonstrated that the model achieved favorable overall detection performance across different defect types while maintaining satisfactory inference efficiency. The proposed surface defect detection method for steel containers improves defect recognition in complex industrial scenarios and achieves a favorable balance between detection accuracy and inference efficiency, providing a technical reference for intelligent surface defect inspection of steel containers and quality control of packaging containers.
关键词
钢容器 /
表面缺陷 /
视觉检测 /
Mamba-YOLO
Key words
steel container /
surface defect detection /
visual detection /
Mamba-YOLO
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] DESHWAL G K, PANJAGARI N R.Review on Metal Packaging: Materials, Forms, Food Applications, Safety and Recyclability[J]. Journal of Food Science and Technology, 2020, 57(7): 2377-2392.
[2] TANG B, CHEN L, SUN W, et al.Review of Surface Defect Detection of Steel Products Based on Machine Vision[J]. IET Image Processing, 2023, 17(2): 303-322.
[3] REN S, HE K, GIRSHICK R, et al.Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. Advances in Neural Information Processing Systems, 2015, 28.
[4] REDMON J, DIVVALA S, GIRSHICK R, et al.You Only Look Once: Unified, Real-Time Object Detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 779-788.
[5] VASWANI A, SHAZEER N, PARMAR N, et al.Attention Is All You Need[J]. Advances in Neural Information Processing Systems, 2017, 30.
[6] ZHU L, LIAO B, ZHANG Q, et al. Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model[J]. arXiv Preprint arXiv:2401.09417, 2024.
[7] WANG Z, LI C, XU H, et al.Mamba YOLO: A Simple Baseline for Object Detection with State Space Model[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2025, 39(8): 8205-8213.
[8] SONG K, YAN Y.A Noise Robust Method Based on Completed Local Binary Patterns for Hot-Rolled Steel Strip Surface Defects[J]. Applied Surface Science, 2013, 285: 858-864.
[9] XIE X.A Review of Recent Advances in Surface Defect Detection Using Texture Analysis Techniques[J]. ELCVIA: Electronic Letters on Computer Vision and Image Analysis, 2008: 1-22.
[10] 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.
[11] YI H, LIU B, ZHAO B, et al.Small Object Detection Algorithm Based on Improved YOLOv8 for Remote Sensing[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 17: 1734-1747.
[12] GIRI K J.SO-YOLOv8: A Novel Deep Learning-Based Approach for Small Object Detection with YOLO Beyond COCO[J]. Expert Systems with Applications, 2025, 280: 127447.
[13] SHIN J H, HAMZA M, KIM H, et al.Lightweight Swin Transformer for High-Precision Industrial Defect Detection in Smart Manufacturing[J]. Alexandria Engineering Journal, 2025, 130: 227-240.
[14] AYON S T K, SIRAJ F M, UDDIN J. Steel Surface Defect Detection Using Learnable Memory Vision Transformer[J]. Computers, Materials & Continua, 2025, 82(1).
[15] BAO M, LYU S, XU Z, et al.Vision Mamba in Remote Sensing: A Comprehensive Survey of Techniques, Applications and Outlook[J]. Remote Sensing, 2026, 18(4): 594.
基金
国家磁约束核聚变能发展研究专项课题（No.2022YFE03170001）
PDF(1603 KB)
渝公网安备 50010702501716号 渝ICP备15012534号-2
