基于改进DeepLabv3+模型的电阻点焊表面缺陷检测方法

doi:10.3969/j.issn.1004-132X.2026.09.025

摘要/Abstract

摘要：

针对机器视觉技术对电阻点焊表面的油污、反光、曲面等因素敏感，使得图像处理困难和分割精度低等问题，提出一种基于深度学习的语义分割方法，通过改进DeepLabv3+模型实现对焊点图像特征的智能分割、识别与分类。实验结果表明：采用MobileNetV3-Large作为主干网络实现了模型轻量化；添加改进的CBAM-change注意力模块提高了模型的分割性能；改进后的DeepLabv3+模型的平均交并比达到92.15%、F1分数达到95.86%、准确率达到99.81%、处理单张焊点图像的平均时间为26.24 ms，可实现白车身焊点缺陷精准分割。

关键词: 电阻点焊, 语义分割, DeepLabv3+模型, 缺陷检测

Abstract:

Aiming at the issues that machine vision technology was sensitive to factors such as oil stains， reflections， and curved surfaces on resistance spot welding surfaces， making image processing difficult and segmentation accuracy low， a semantic segmentation method was proposed based on deep learning. By improving the DeepLabv3+ model， intelligent segmentation， recognition， and classification of welding spot image features were achieved. Experimental results show that： using MobileNetV3-Large as the backbone network model lightweight is achieved. Adding an improved CBAM-change attention module is the model's segmentation performance is further enhanced. The improved DeepLabv3+ model achieves a mean intersection over union of 92.15%， an F1 score of 95.86%， an accuracy of 99.81%， and an average processing time of 26.24 ms per welding spot image， enabling precise segmentation of white body welding spot defects.

Key words: resistance spot welding, semantic segmentation, DeepLabv3+model, defect detection

中图分类号:

TG438

熊鑫, 吴玮, 杜洁, 向词裴. 基于改进DeepLabv3+模型的电阻点焊表面缺陷检测方法[J]. 中国机械工程, 2026, 37(9): 2326-2333.

XIONG Xin, WU Wei, DU Jie, XIANG Cipei. A Method for Surface Defect Detection in Resistance Spot Welding Based on an Improved DeepLabv3+ Model[J]. China Mechanical Engineering, 2026, 37(9): 2326-2333.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.cmemo.org.cn/CN/10.3969/j.issn.1004-132X.2026.09.025

https://www.cmemo.org.cn/CN/Y2026/V37/I9/2326

图/表 16

图1 技术路线图

Fig.1 Technology roadmap

图2 实验设备

Fig.2 Experimental equipment

表1 实验工艺参数

Tab.1 Experimental process parameters

参数名称	数值范围
电流/kA	6~12
时间/ms	100~300
压力/kN	5~7

图3 焊点图像数据集

Fig.3 Welding spot image dataset

图4 DeepLabv3+网络结构［14］

Fig.4 DeepLabv3+ network architecture［14］

图5 改进后的DeepLabv3+网络结构

Fig.5 Improved DeepLabv3+ network structure

图6 MobileNetV3结构

Fig.6 MobileNetV3 structure

图7 CBAM注意力模块

Fig.7 Convolutional block attention module

图8 CAM和SAM注意力模块

Fig.8 Channel attention module and spatial attention module

图9 激活函数对比

Fig.9 Comparison of activation functions

图10 损失曲线及mIoU曲线

Fig.10 Loss curve and mIoU curve

表2 模型改进结果对比

Tab.2 Comparison results of model improvement

模型		I_mIoU/%	F1分数/%	准确率A/%	时间t/ms
1	图10a	91.79	95.64	99.80	31.37
2	图10b	87.79	93.99	99.70	21.07
3	图10c	90.17	94.82	99.77	26.09
4	图10d	91.06	95.26	99.78	26.38
5	图10e	92.15	95.86	99.81	26.24

图11 不同模型可视化

Fig.11 Visualization of different models

表3 不同模型结果对比

Tab.3 Comparison of results of different models

模型	I_mIoU/%	F1分数/%	准确率A/%	参数量/10⁶	GFLOPs
U-Net	92.13	95.89	99.82	43.934	184.737
PSPNet	88.53	93.82	99.70	2.377	6.034
HrNet	92.72	96.17	99.83	9.642	37.487
DeepLabv3+（Ours）	92.15	95.86	99.81	4.923	16.700

图12 RSW-D的损失曲线及mIoU曲线

Fig.12 Loss curve and mIoU curve of RSW-D

图13 RSW-D可视化

Fig.13 Visualization of RSW-D

参考文献 23

[1]	CHUENMEE N， PHOTHI N， CHAMNIPRASART K， et al. Machine Learning for Predicting Resistance Spot Weld Quality in Automotive Manufacturing［J］. Results in Engineering， 2025， 25： 103570.
[2]	XIAO Meng， YANG Bo， WANG Shilong， et al. A Feature Fusion Enhanced Multiscale CNN with Attention Mechanism for Spot-welding Surface Appearance Recognition［J］. Computers in Industry， 2022， 135： 103583.
[3]	XIAO Meng， YANG Bo， WANG Shilong， et al. Fine Coordinate Attention for Surface Defect Detection［J］. Engineering Applications of Artificial Intelligence， 2023， 123： 106368.
[4]	LONG J， SHELHAMER E， DARRELL T. Fully Convolutional Networks for Semantic Segmentation［C］∥2015 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE， 2015： 3431-3440.
[5]	XIAO Xiaoyang， ZHAO Yuqian， ZHANG Fan， et al. BASeg： Boundary Aware Semantic Segmentation for Autonomous Driving［J］. Neural Networks， 2023， 157： 460-470.
[6]	LIU Yang， WANG Yongfu， LI Qiansheng. Lane Detection Based on Real-time Semantic Segmentation for End-to-end Autonomous Driving under Low-light Conditions［J］. Digital Signal Processing， 2024， 155： 104752.
[7]	WEI Kan， DAI Jinkun， HONG Danfeng， et al. MGFNet： an MLP-dominated Gated Fusion Network for Semantic Segmentation of High-resolution Multi-modal Remote Sensing Images［J］. International Journal of Applied Earth Observation and Geoinformation， 2024， 135： 104241.
[8]	GONG Zhi， DUAN Lijuan， XIAO Fengjin， et al. MSAug： Multi-strategy Augmentation for Rare Classes in Semantic Segmentation of Remote Sensing Images［J］. Displays， 2024， 84： 102779.
[9]	WANG Haonan， CAO Peng， YANG Jinzhu， et al. Narrowing the Semantic Gaps in U-Net with Learnable Skip Connections： The Case of Medical Image Segmentation［J］. Neural Networks， 2024， 178： 106546.
[10]	LIU Xiaoxiao， ZHAO Yan， WANG Shigang， et al. Mambav3d： a Mamba-based Virtual 3D Module Stringing Semantic Information between Layers of Medical Image Slices［J］. Displays， 2024， 85： 102890.
[11]	SUN Haohan， XIA Liqian， ZHOU Yihu， et al. Online Detection and Evaluation of Weld Surface Defects Based on Lightweight Network VGG16-UNet and Laser Scanning［J］. Journal of Manufacturing Processes， 2024， 129： 292-306.
[12]	XU H， YAN Z H， JI B W， et al. Defect Detection in Welding Radiographic Images Based on Semantic Segmentation Methods［J］. Measurement， 2022， 188： 110569.
[13]	REDMON J， FARHADI A. Yolov3： an Incremental Improvement［J］. Arxiv Preprint Arxiv，2018：1804.02767 .
[14]	CHEN L C， ZHU Yukun， PAPANDREOU G， et al. Encoder-decoder with Atrous Separable Convolution for Semantic Image Segmentation［C］∥Computer Vision – ECCV 2018. Cham： Springer， 2018： 833-851.
[15]	CHOLLET F. Xception： Deep Learning with Depthwise Separable Convolutions［C］∥2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE， 2017： 1800-1807.
[16]	SANDLER M， HOWARD A， ZHU Menglong， et al. MobileNetV2： Inverted Residuals and Linear Bottlenecks［C］∥2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE， 2018： 4510-4520.
[17]	HOWARD A G， ZHU M， CHEN B， et al. MobileNets： Efficient Convolutional Neural Networks for Mobile Vision Applications［J］. Arxiv Preprint Arxiv， 2017：1704.04861.
[18]	HOWARD A， SANDLER M， CHEN Bo， et al. Searching for MobileNetV3［C］∥2019 IEEE/CVF International Conference on Computer Vision （ICCV）. IEEE， 2019： 1314-1324.
[19]	HU Jie， SHEN Li， SUN Gang. Squeeze-and-excitation Networks［C］∥2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE， 2018： 7132-7141.
[20]	WOO S， PARK J， LEE J Y， et al. CBAM： Convolutional Block Attention Module［C］∥Computer Vision-ECCV 2018. Cham： Springer， 2018： 3-19.
[21]	RONNEBERGER O， FISCHER P， BROX T. U-Net： Convolutional Networks for Biomedical Image Segmentation［C］∥Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. Cham： Springer， 2015： 234-241.
[22]	ZHAO Hengshuang， SHI Jianping， QI Xiaojuan， et al. Pyramid Scene Parsing Network［C］∥2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE， 2017： 6230-6239.
[23]	SUN Ke， ZHAO Yang， JIANG Borui， et al. High-resolution Representations for Labeling Pixels and Regions［EB/OL］. 2019： arXiv： 1904.04514.