多工件拼接焊缝面结构光滑度的视觉检测技术研究  

Research on Visual Inspection Technology for Structural Smoothness of Multi Workpiece Splicing Weld Surface

作　　者：何涛 HE Tao(School of Information Engineering and Technology,Changzhou Vocational Institute of Industry Technology,Changzhou 2130o0,China)

机构地区：[1]常州工业职业技术学院信息工程学院,江苏常州213000

出　　处：《计算机测量与控制》2023年第12期90-96,166,共8页Computer Measurement &Control

基　　金：江苏省自然科学基金(202109874660)。

摘　　要：由于多工件拼接焊缝面结构光滑度检测过程中,受其拼接面复杂且图像存在噪声,导致细节特征不明显,从而使检测精度、效率降低,为此,提出多工件拼接焊缝面结构光滑度的视觉检测技术;采用字典学习方法,对焊缝面图像去噪处理,将焊缝图像去噪问题转化为最小化问题,消除焊缝面图像中存在的噪声,将其输入MRFENet网络,结合深度学习和残差学习技术,提取图像特征,采用逐级特征融合方法,融合提取的深层特征和浅层特征,实现焊缝面图像的增强处理;采用增量二维主成分分析提取焊缝面结构的光滑度特征,结合焊缝面图像的协方差矩阵更新焊缝图像特征矩阵,结合残差函数,计算焊缝面结构高阶光滑度特征向量,实现光滑度检测;实验结果表明,所提方法图像处理效果好、检测精度高,且一直处于94%以上,检测效率高。Due to the complexity of the joint surface and the noise of the image during the smoothness detection process of multi-workpiece joint weld surface,the detail features are not obvious,resulting in the decreases of detection accuracy and efficiency.Therefore,a visual detection technology of multi-workpiece joint weld surface structure smoothness is proposed.The dictionary learn-ing method is adopted to de-noise the weld surface image,transform the denoising problem of the weld surface image into a minimiza-tion problem,eliminate the noise of the weld surface image,input it into the MRFENet network,combine the deep learning and resid-ual learning technology,extract the image features,adopt the step-by-step feature fusion method to integrate the extracted deep fea-tures and shallow features,and realize the enhancement processing of the weld surface image;The incremental two-dimensional prin-cipal component analysis is used to extract the smoothness features of the weld surface structure;The covariance matrix of the weld surface image is combined to update the weld image feature matrix,and combined with the residual function,the high-order smooth-ness feature vector of the weld surface structure is calculated to realize the smoothness detection.The experimental results show that the proposed method has a good image processing effect and high detection accuracy,the detection accuracy is always above 94%,which has a high detection efficiency.

关 键 词：焊缝面结构 字典学习 视觉检测技术 特征提取 光滑度检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]