基于DETR网络的小样本铸件飞边检测研究
Fringe Detection of Few-shot Casting Based on DETR
- 2025年45卷第2期 页码:187-192
收稿日期:2024-01-04,
修回日期:2024-02-07,
录用日期:2024-02-28,
纸质出版日期:2025-02-20
DOI: 10.15980/j.tzzz.T20240002
移动端阅览
浏览全部资源
扫码关注微信
收稿日期:2024-01-04,
修回日期:2024-02-07,
录用日期:2024-02-28,
纸质出版日期:2025-02-20
移动端阅览
王梦雪,穆春阳,马行,等. 基于DETR网络的小样本铸件飞边检测研究[J]. 特种铸造及有色合金,2025,45(2):187-192.
WANG M X,MU C Y,MA X,et al. Fringe detection of few-shot casting based on DETR[J]. Special Casting & Nonferrous Alloys,2025,45(2):187-192.
为实现铸件飞边自动化检测,针对检测过程中铸件飞边样本量少,模型特征融合不充分的问题,基于DETR网络提出一种基于注意力加权特征融合的小样本目标检测算法。首先,设计坐标注意力加权的金字塔特征融合网络,以提高模型特征信息交互融合能力和感兴趣区域专注力;然后引入二维图像相对位置编码,增强对平移不变性的模式识别能力;后处理阶段引入Smooth-L1优化损失函数,提高模型检测精度。结果表明,改进算法在试验实例为3、5、10和30个时,检测精度分别为47.50%、54.06%、66.00%和79.32%,改进算法在自制铸件飞边数据集上的检测精确度更高。
In view of small amounts of specimens of casting fringe during detection process and the insufficient fusion of model feature, a few-shot target detection algorithm characterzied by attention-weighted feature fusion was proposed based on DETR network to realize the automatic detection of casting fringe. Firstly, a pyramid feature fusion network with coordinate attention weighting was designed to improve interactive fusion capability and interest region focus of model feature information. Then, the relative position coding of 2D image was introduced to enhance the pattern recognition ability for translation invariance. Smooth-L1 optimized loss function was introduced in the post-processing stage to improve detection accuracy. The results indicate that the detection accuracy of modified algorithm is 47.50%, 54.06%, 66.00%, and 79.32% in test case of 3, 5, 10, and 30, respectively, indicating higher detection accuracy of modified algorithm on the homemade casting fretting dataset.
冯纵 , 陈新度 , 吴磊 , 等 . 基于余弦相似度的铸件浇冒口检测研究 [J]. 特种铸造及有色合金 , 2021 , 41 ( 8 ): 952 - 956 .
吴鲁纪 , 秦佳音 , 李安虎 , 等 . 机器视觉识别技术在机械传动领域的发展与应用 [J]. 机械传动 , 2022 , 46 ( 7 ): 167 - 176 .
曹家乐 , 李亚利 , 孙汉卿 , 等 . 基于深度学习的视觉目标检测技术综述 [J]. 中国图象图形学报 , 2022 , 27 ( 6 ): 1 697 - 1 722 .
JIANG W , HUANG K , GENG J , et al . Multi-scale metric learning for few-shot learning [J]. IEEE Transactions on Circuits and Systems for Video Technology , 2020 , 31 ( 3 ): 1 091 - 1 102 .
HUANG G , LARADJI I , VAZQUEZ D , et al . A survey of self-supervised and few-shot object detection [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2022 , 45 ( 4 ): 4 071 - 4 089 .
FENG H T , ZHANG L , YANG X , et al . Incremental few-shot object detection via knowledge transfer [J]. Pattern Recognition Letters , 2022 , 156 : 67 - 73 .
LI S M , XUE L S , FENG L , et al . Hybrid convolutional-transformer framework for drone-based few-shot weakly supervised object detection [J]. Computers and Electrical Engineering , 2022 , 102 : 108 154 .
郭保青 , 张德芬 . 基于度量元学习的铁路小样本入侵目标检测方法 [J]. 光学精密工程 , 2023 , 31 ( 12 ): 1 816 - 1 826 .
JOSEPH K J , RAJASEGARAN J , KHAN S , et al . Incremental object detection via meta-learning [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2022 , 44 ( 12 ): 9 209 - 9 216 .
ZHANG G J , CUI K W , WU R L , et al . PNPDet: Efficient few-shot detection without forgetting via plug-and-play sub-networks [A]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision [C]. Waikoloa, USA , 2021 .
杜芸彦 , 李鸿 , 杨锦辉 , 等 . 基于负边距损失的小样本目标检测 [J]. 计算机应用 , 2022 , 42 ( 11 ): 3 617 - 3 624 .
LI A X , LI Z G . Transformation invariant few-shot object detection [A]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition [C]. Nashville, USA , 2021 .
窦智 , 胡晨光 , 李庆华 , 等 . 改进YOLOv7的小样本钢板表面缺陷检测算法 [J]. 计算机工程与应用 , 2023 , 59 ( 23 ): 283 - 292 .
CARION N , NASSA F , SYNNAEVE G , et al . End-to-end object detection with transformers [A]// European Conference on Computer Vision [C]. Cham .: Springer International Publishing , 2020 .
HOU Q B , ZHOU D Q , FENG J S . Coordinate attention for efficient mobile network design [A]// Proceeding of the IEEE/CVF Conference on Computer Vision and Pattern Recognition [C]. Nashville, USA , 2021 .
相关作者
相关机构