[1]黄 宇*,陈建国,严 南.doi: 10.3969/j.issn.1001-3849.2025.09.005目标区域标定下金属工件电镀刷表面缺陷识别研究[J].电镀与精饰,2025,(09):29-37.
　Huang Yu*,Chen Jianguo,Yan Nan.Research on surface defect recognition of metal workpiece electroplating brush under target area calibration[J].Plating & Finishing,2025,(09):29-37.
点击复制

doi: 10.3969/j.issn.1001-3849.2025.09.005目标区域标定下金属工件电镀刷表面缺陷识别研究()

《电镀与精饰》[ISSN:1001-3849/CN:12-1096/TG]

卷:
期数:
2025年09
页码:
29-37
栏目:
出版日期:
2025-09-30

文章信息/Info

Title:
Research on surface defect recognition of metal workpiece electroplating brush under target area calibration
作者:
黄 宇*; 陈建国; 严 南
(成都理工大学 工程技术学院,四川 乐山 614000)
Author(s):
Huang Yu*;  Chen Jianguo;  Yan Nan
(The Engineering & Technical College of Chengdu University of Technology, Leshan 614000, China)
关键词:
目标区域标定; 金属工件; 电镀刷; 表面缺陷识别; YOLOv8算法
Keywords:
target area calibration;  metal workpieces;  electroplating brush;  surface defect identification;  yolov8 algorithm
分类号:
TP391.41
文献标志码:
A
摘要:
金属工件电镀刷表面缺陷识别过程中，因目标尺度小、排列密集和背景干扰使得缺陷目标区域难以判定，特别是当缺陷目标区域的特征不明显或与其他表面特征相混淆时，从目标区域提取特征的难度进一步加大，增加了缺陷识别的整体难度。为此，提出一种基于目标区域标定的金属工件电镀刷表面缺陷识别方法。该方法通过确定金属工件电镀刷表面的方差纹理图，并根据平均标准方差与平均偏移量设定阈值，对方差纹理图展开分割，从复杂的图像中初步筛选出存在缺陷的目标区域。引入PConv改进Black网络，以此增强网络对不规则区域和缺失数据的处理能力，使得网络能够更有效地从目标区域中提取特征。通过洗牌注意力机制(SA)对Neck网络进行优化，实现特征的有效融合。以融合后的特征为基础，改进YOLOv8算法通过计算交并比和损失函数，实现对金属工件电镀刷表面缺陷的准确识别。实验结果表明：该方法具有较高的目标提取精度与缺陷识别精度，误识率相对较低，在实际应用中展现出良好的效果。
Abstract:
In the process of surface defect identification for electroplated brushes on metal workpieces, the identification of defective target regions is challenging due to the small scale of the targets, their dense arrangement, and background interference. In particular, when the characteristics of defective target regions are inconspicuous or confused with other surface features, the difficulty of extracting features from these regions is further exacerbated, thereby increasing the overall complexity of defect recognition. To address this issue, a surface defect identification method for electroplated brushes on metal workpieces based on target region calibration is proposed. This method involves determining the variance texture map of the electroplated brush surface on metal workpieces and setting thresholds based on the average standard deviation and average offset to segment the variance texture map, thereby preliminarily isolating defective target regions from complex images. The Black network is enhanced by incorporating PConv to improve the network’s capability in handling irregular regions and missing data, enabling more effective feature extraction from the target regions. The Neck network is optimized using the shuffle attention (SA) mechanism to achieve efficient feature fusion. Based on the fused features, the YOLOv8 algorithm is improved to accurately identify surface defects on electroplated brushes of metal workpieces by calculating the intersection over union ratio and loss function. Experimental results demonstrate that the proposed method achieves high precision in both target extraction and defect identification, with a relatively low false recognition rate, exhibiting promising performance in practical applications.

参考文献/References:

[1].于文彬, 皮志超, 肖鹏, 等. 钢铁表面电镀刷金属及其性能研究进展[J]. 机械制造与自动化, 2024, 53(4): 7-19, 36.
[2].宿磊, 王立建, 祁阳, 等. 基于IADSA深度迁移网络的金属表面缺陷检测[J]. 机械工程学报, 2023, 59(24): 46-55.
[3].Wang Y, Zhang C, Li H, et al. Identification and extraction of surface defects on composite workpieces based on Multilayer Perceptron‐Moth Flame algorithm two‐stage neural network[J]. Polymer Composites, 2024, 45(3): 2725-2738.
[4].王一, 龚肖杰, 程佳. 基于改进U-net的金属工件表面缺陷分割方法[J]. 激光与光电子学进展, 2023, 60(15): 333-338.
[5].Xu Z, Fang P, Yang X, et al. YOLOv7-Rep: A re-parameterization method for surface defect detection in workpieces[J]. Journal of Real-Time Image Processing, 2025, 22(1): 1-13.
[6].王传钊, 姜秀海, 晁永生, 等. 基于涡流脉冲热成像的焊缝表面多缺陷检测[J]. 红外技术, 2023, 45(1): 84-90.
[7].林煜桐, 朱姗姗, 彭凌西, 等. 基于局部方差和后验概率分类的快速模板匹配算法[J]. 电子技术应用, 2023, 46(9): 97-102.
[8].王超, 郭恒光. 基于方向—频率分解的纹理特征提取方法研究[J]. 软件, 2024, 45(4): 150-153.
[9].谢竞, 邓月明, 王润民. 改进YOLOv8s的交通标志检测算法[J]. 计算机工程, 2024, 50(11): 338-349.
[10].赵志宏, 郝子晔. 改进YOLOv8的航拍小目标检测方法:CRP-YOLO[J]. 计算机工程与应用, 2024, 60(13): 209-218.
[11].李希, 潘誉. 基于改进YOLOv8的探地雷达管线目标检测方法[J]. 计算机与现代化, 2025(1): 94-99.
[12].张雷, 武文喆, 白雪媛. 基于Light-YOLOv8的围棋棋谱识别[J]. 计算机科学, 2024, 51(S2): 347-353.
[13].吴永泽, 俞建峰, 化春键, 等. 基于改进YOLOv8n的圆柱电池壳多维缺陷检测技术研究[J]. 电子测量与仪器学报, 2024, 38(12): 62-71.
[14].王礼乐, 刘渊. 基于空间注意力图的知识蒸馏算法[J]. 计算机应用研究, 2024, 41(6): 1693-1698.
[15].严炳杨, 冯志荣, 鲍灵辉, 等. DAR-Net:双空间注意力与边界细化的遥感影像滑坡提取网络[J]. 地理空间信息, 2025, 23(5): 1-4, 38.
[16].贾军, 任祺. 基于改进YOLOv8的飞机检测研究[J]. 工业控制计算机, 2024, 37(9): 9-11.
[17].王莹, 张上, 胡益民, 等. 改进YOLOv8的遥感图像飞机目标轻量化检测算法[J]. 遥感信息, 2025, 40(1): 134-141.
[18].邱琳琳, 朱卫纲, 李永刚, 等. 基于改进YOLOv8的SAR图像飞机目标检测算法[J]. 电光与控制, 2025, 32(3): 101-110.
[19].蔡引娣, 张殿鹏, 孙梓盟, 等. 基于改进YOLOv8模型的增材制造微小气孔缺陷检测及其尺寸测量[J]. 光学精密工程, 2024, 32(21): 3222-3230.

更新日期/Last Update: 2025-09-11