基于BP-DFO Faster R-CNN的工业产品小样本表面缺陷检测模型

doi:10.16731/j.cnki.1671-3133.2025.12.001

现代制造工程 ›› 2025, Vol. 543 ›› Issue (12): 1-10.doi: 10.16731/j.cnki.1671-3133.2025.12.001

• 试验研究 • 下一篇

基于BP-DFO Faster R-CNN的工业产品小样本表面缺陷检测模型

贺永姣^1,2, 王林², 章晓爽^1,2, 黎格献^1,2, 杜阳^1,2

1 贵州民族大学数据科学与信息工程学院,贵阳 550025;
2 贵州民族大学贵州省模式识别与智能系统重点实验室,贵阳 550025

收稿日期:2025-02-20 出版日期:2025-12-18 发布日期:2026-01-06
通讯作者: 王林,博士,教授,硕导,主要研究方向为图像处理与机器视觉。E-mail:wanglin@gzmu.edu.cn
作者简介:贺永姣,硕士研究生,主要研究方向为工业图像缺陷检测。章晓爽,硕士研究生,主要研究方向为面部表情识别。黎格献,硕士研究生,主要研究方向为小样本图像分类。杜阳,硕士研究生,主要研究方向为工业缺陷检测。E-mail:1991424263@qq.com
基金资助:
*国家自然科学基金项目(62276103);贵州省科技计划项目(黔科合基础-ZK[2022]一般195,黔科合平台人才-ZCKJ[2021]007);贵州省青年科技人才成长项目(黔教合KY字[2021]104);贵州省模式识别与智能系统重点实验室开放课题项目(GZMUKL[2022]KF01,GZMUKL[2022]KF05);贵州民族大学基金科研项目(GZMUZK[2023]YB14,黔教技[2024]063号)

Industrial product surface defect detection model based on BP-DFO Faster R-CNN for few-shot

HE Yongjiao^1,2, WANG Lin², ZHANG Xiaoshuang^1,2, LI Gexian^1,2, DU Yang^1,2

1 School of Data Science and Information Engineering,Guizhou Minzu University,Guiyang 550025,China;
2 Key Laboratory of Pattern Recognition and Intelligent System,Guizhou Minzu University, Guiyang 550025,China

Received:2025-02-20 Online:2025-12-18 Published:2026-01-06

摘要/Abstract

摘要： 在现代工业产品的生产中,目标缺陷检测技术面临诸多挑战,如提取的共享特征不平衡、特征通道权重分配不精确和缺陷检测精度较低等问题。针对上述问题,提出一种BP-DFO Faster R-CNN缺陷检测模型。该模型在目标金字塔结构后引入平衡金字塔机制,并在此基础上增加动态特征优化架构。不仅解决了共享特征不平衡的问题,还实现了全局上下文信息和局部细节特征的高效融合,从而克服传统通道注意力机制因忽视局部信息而导致的特征权重分配不准确的问题。最后,对区域生成网络(Region Proposal Network,RPN)进行了改进,在其头部集成轻量级通道优化器,显著增强区域生成网络对不同特征和尺度通道的选择能力,从而优化缺陷检测任务的性能。在NEU-DET、PCB、GC10-DET缺陷数据集上进行对比实验,在5-way 5-shot的训练策略中分别取得了31.0 %、22.1 %和21.8 %的平均精度,在5-way 10-shot的训练策略中分别取得了50.2 %、35.5 %和31.5 %的平均精度,均优于其他对比模型。

关键词: 工业产品, 表面缺陷检测, 平衡金字塔机制, Faster R-CNN, 动态特征优化, 小样本

Abstract: In modern industrial production,object defect detection faced numerous challenges,such as imbalanced shared feature extraction,inaccurate channel weight assignment,and low detection accuracy. To address these issues,a novel defect detection model BP-DFO Faster R-CNN was proposed. The model introduced a balanced feature pyramid mechanism after the feature pyramid network and incorporated a dynamic feature optimization architecture architecture. This design not only alleviated the imbalance in shared features but also enabled effective fusion of global contextual information and local detailed features,thereby overcoming the inaccurate feature weighting caused by traditional channel attention mechanisms that neglected local details. Furthermore,the Region Proposal Network (RPN) was improved by integrating a lightweight channel optimizer into its head,which significantly enhanced the RPN′s ability to select features across different channels and scales,thus optimizing performance in defect detection tasks. Extensive experiments were conducted on the NEU-DET,PCB,and GC10-DET defect datasets.Under the 5-way 5-shot training strategy,the model achieved mean average precisions of 31.0 %,22.1 %,and 21.8 %,respectively. Under the 5-way 10-shot setting,it achieved 50.2 %,35.5 %,and 31.5 %,outperforming all compared models.

Key words: industrial products, surface defect detection, balanced pyramid scheme, Faster R-CNN, dynamic feature optimization, few-shot

中图分类号:

TP391.41

贺永姣, 王林, 章晓爽, 黎格献, 杜阳. 基于BP-DFO Faster R-CNN的工业产品小样本表面缺陷检测模型[J]. 现代制造工程, 2025, 543(12): 1-10.

HE Yongjiao, WANG Lin, ZHANG Xiaoshuang, LI Gexian, DU Yang. Industrial product surface defect detection model based on BP-DFO Faster R-CNN for few-shot[J]. Modern Manufacturing Engineering, 2025, 543(12): 1-10.

参考文献

[1] 杨泽青,张明轩,陈英姝,等.基于机器视觉的表面缺陷检测方法研究进展[J].现代制造工程,2023(4):143-156.
[2] 史燕燕,史殿习,乔子腾,等.小样本目标检测研究综述[J].计算机学报,2023,46(8):1753-1780.
[3] 赵凯琳,靳小龙,王元卓.小样本学习研究综述[J].软件学报,2020,32(2):349-369.
[4] KÖHLER M,EISENBACH M,GROSS H M.Few-shot object detection:a comprehensive survey[J].IEEE Transactions on Neural Networks and Learning Systems,2023,35(9):11958-11978.
[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,2016,39(6):1137-1149.
[6] LIU W,ANGUELOV D,ERHAN D,et al. SSD:Single Shot multibox Detector[C]//Computer Vision-ECCV 2016:14 th European Conference. Amsterdam: Springer International Publishing,2016:21-37.
[7] CHEN H,WANG Y,WANG G,et al. LSTD:a low-shot transfer detector for object detection[C]//Proceedings of the AAAI conference on artificial intelligence.[S.l.]:AAAI Press,2018,32:2836-2843.
[8] ZHANG G,CUI K,WU R,et al. PNPDet:efficient few-shot detection without forgetting via plug-and-play sub-networks[C]//CVF Winter Conference on Applications of Computer Vision.[S.l.]:IEEE,2021:3823-3832.
[9] YAN X,CHEN Z,XU A,et al. Meta R-CNN:towards general solver for instance-level low-shot learning[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul:IEEE,2019:9576-9585.
[10] WU A,HAN Y,ZHU L,et al. Universal-prototype enhancing for few-shot object detection[C]//2021 IEEE/CVF International Conference on Computer Vision.[S.l.]:IEEE,2021:9567-9576.
[11] AZEEM A,LI Z,SIDDIQUE A,et al.Unified multimodal fusion transformer for few shot object detection for remote sensing images[J].Information Fusion,2024,111:102508.
[12] SUN E,ZHOU D,TIAN Y,et al. Transformer-based few-shot object detection in traffic scenarios[J].Applied Intelligence,2024,54(1):947-958.
[13] LIN T Y,DOLLÁR P,GIRSHICK R,et al.Feature pyramid networks for object detection[C]//2017 IEEE/CVF conference on Computer Vision and Pattern Recognition.Honolulu:IEEE,2017:936-944.
[14] LIU S,QI L,QIN H,et al. Path aggregation network for instance segmentation[C]//CVF conference on Computer Vision and Pattern Recognition.Salt City:IEEE,2018:8759-8768.
[15] GHIAISG,LIN T Y,LE Q V. NAS-FPN:learning scalable feature pyramid architecture for object detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.[S.l.]:IEEE,2019:7036-7045.
[16] WANG H,LI Y,WANG S. CRPN:distinguish novel categories via class-relevant region proposal network for few-shot object detection[C]//ICASSP 2022-2022 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP). Singapore:IEEE,2022:2230-2234.
[17] 宋雨,李敏,何玉杰,等.一种新的强分辨性零样本目标检测方法[J].计算机应用研究,2022,39(11):3475-3480,3515.
[18] 潘雪玲,李国和,郑艺峰.面向深度网络的小样本学习综述[J].计算机应用研究,2023,40(10):2881-2888,2895.
[19] SUN B,LI B,CAI S,et al. FSCE:few-shot object detection via contrastive proposal encoding[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recogni-tion.Nashville:IEEE,2021:7352-7362.
[20] WANG X,HUANG T E,DARRELL T,et al.Frustratingly simple few-shot object detection[J].arXiv preprint arXiv:2003.06957,2020.
[21] WU J X,LIU S T,HUANG I,et al. Multi-scale positive sample refinement for few-shot object detection[J]. arXiv preprint arXiv,2007.09384,2020.
[22] HAN J,REN Y,DING J,et al. Few-shot object detection via variational feature aggregation[C]//Proceedings of the AAAI Conference on Artificial Intelligence.Washington:AAAI,2023,37(1):755-763.
[23] XIN Z M,WU T X,CHEN S M,et al. ECEA:Extensible Co-existing Attention for few-shot object detection[J].IEEE Transactions on Image Processing,2024,33:5564-5576.
[24] HU H,BAI S,LI A,et al. Dense relation distillation with context-aware aggregation for few-shot object detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Nashville:IEEE,2021:10180-10189.
[25] PANG J,CHEN K,SHI J,et al. Libra R-CNN:towards balanced learning for object detection[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Long beach:IEEE,2019:821-830.
[26] OORD A,LI Y,VINYALS O. Representation learning with contrastive predictive coding[J].arXiv preprint arXiv,1807.03748,2018.
[27] BAO Y Q,SONG K C,LIU J,et al. Triplet-graph reasoning network for few-shot metal generic surface defect segmentation[J]. IEEE Transactions on Instrumentation and Measuremente,2021,70:1-11.
[28] HUANG W,WEI P A. A PCB dataset for defects detection and classification[J].Arxiv preprint arxiv,1901.08204, 2018.
[29] LV X,DUAN F,JIANG J,et al. Deep metallic surface defect detection:the new benchmark and detection network[J].Sensors,2020,20(6):1562.
[30] FAN Q,ZHUO W,TANG C K,et al. Few-shot object detection with attention-RPN and multi-relation detector[C]//2020 IEEE/CVF conference on computer vision and pattern recogni-tion. Seattle:IEEE,2020:4012-4021.
[31] YAN X,CHEN Z,XU A,et al. Meta R-CNN:towards general solver for instance-level low-shot learning[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul:IEEE,2019:9576-9585.

基于BP-DFO Faster R-CNN的工业产品小样本表面缺陷检测模型

Industrial product surface defect detection model based on BP-DFO Faster R-CNN for few-shot

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 7

编辑推荐

Metrics

[1]	李季村, 郑鹏, 李岩, 何青泽. 融合GhostBottleneck及注意力机制的铝型材表面缺陷检测算法研究^*[J]. 现代制造工程, 2025, 534(3): 115-123.
[2]	汪劲雯, 许新科, 赵龙彪, 袁狄剑, 杜昱. 基于像素纹理特征的钢卷端面缺陷检测方法[J]. 现代制造工程, 2025, 543(12): 105-113.
[3]	郭康, 王志刚, 徐增丙. 基于AFF-Stablenet模型的小样本轴承故障诊断^*[J]. 现代制造工程, 2024, 528(9): 144-151.
[4]	孙铁强, 刘俊, 宋超, 肖鹏程. 基于改进YOLOv8n的铝型材表面小目标缺陷检测方法^*[J]. 现代制造工程, 2024, 531(12): 120-129.
[5]	阎馨，杨月川，屠乃威. 基于改进SSD的钢材表面缺陷检测[J]. 现代制造工程, 2023, 512(5): 112-120.
[6]	杨泽青;张明轩;陈英姝;平恩旭;方勇;吕雅丽;高岩. 基于机器视觉的表面缺陷检测方法研究进展[J]. 现代制造工程, 2023, 511(4): 143-156.
[7]	易可夫，胡锦，崔凯阳，胡宏伟. 基于二次特征筛选的覆铜板表面缺陷检测[J]. 现代制造工程, 2021, 489(6): 95-100.