Aiming at the harsh environment and serious light pollution in the production workshop of automobile body-in-white, it is difficult to accurately locate and inefficient when the vision system and other equipment are combined to detect the quality of the solder joints. An improved U-Net image segmentation algorithm was proposed. By improving the convolution structure to better fuse the semantic information of the feature map and lighten the network structure. Improve the loss function and integrate the attention mechanism to better mine the foreground in the case of uneven positive and negative samples, obtain spatial features of different scale feature maps and establish long-term channel relationships. Compared with the original U-Net network, the Dice coefficient of the proposed RPSA-U-Net network is increased by 8.76% to 0.983 6, the MIOU is increased by 11.5% to 0.967 81, and the network parameters are also reduced by 7%. Combined with the image processing method to find the center of the solder joint, the efficiency is higher and the precision is higher, and it has application value.
WANGShuqiang, ZHOUYou, CHENHaolei,et al.Image processing system of welding seam of steel structure based on laser vision[J].Transactions of the China Welding Institution,2022,43(2):101-105.(in Chinese)
HEZhicheng, WANGZhenxing.Welding spot detection method for body in white based on improved YOLOv2[J].Computer Engineering,2020,46(11):246-254. (in Chinese)
HUANGPeng, ZHENGQi, LIANGChao.Overview of image segmentation methods[J].Journal of Wuhan University(Natural Science Edition),2020,66(6):519-531. (in Chinese)
[7]
LONGJ, SHELHAMERE, DARRELLT.Fully convolutional networks for semantic segmentation[C]//CVPR,2015:3431-3440.
[8]
XIAOX, LIANS, LUOZ,et al.Weighted Res-UNet for high-quality retina vessel segmentation[C]//9th International Conference on Information Technology in Medicine and Education(ITME),2018:327-331.
[9]
CHENL C, PAPANDREOUG, KOKKINOSI,et al.Semantic image segmentation with deep convolutional nets and fully connected CRFs[C]//International Conference on Learning Representations,2014:357-361.
[10]
RONNEBERGERO, FISCHERP, BROXT.U-Net:convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention,2015:234-241.
[11]
GUANS, KHANA, SIKDARS,et al.Fully dense UNet for 2D sparse photoacoustic tomography artifact removal[J].IEEE Journal of Biomedical and Health Informatics,2020,24(2):568-576.
[12]
HUANGH, LINL, TONGR,et al.UNet 3+:a full-scale connected UNet for medical image sSegmentation[C]//ICASSP IEEE International Conference on Acoustics,Speech and Signal Processing(ICASSP),IEEE,2020:1055-1059.
[13]
CHENL C, PAPANDREOUG, KOKKINOSI,et al.DeepLab:semantic image segmentation with deep convolutional Nets,Atrous convolution,and fully connected CRFs[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(4):834-848.
LIYuan, LIYanjun, LIUJinchao,et al.Research on segmentation of steel surface defect images based on improved Res-UNet network[J].Journal of Electronics & Information Technology,2022,44(5):1513-1520. (in Chinese)