高级检索
刘天元, 鲍劲松, 汪俊亮, 郑小虎, 王佳铖. 受限解空间下粗粒度正则化的熔池边缘自适应检测方法[J]. 焊接学报, 2020, 41(12): 49-54. DOI: 10.12073/j.hjxb.20200815002
引用本文: 刘天元, 鲍劲松, 汪俊亮, 郑小虎, 王佳铖. 受限解空间下粗粒度正则化的熔池边缘自适应检测方法[J]. 焊接学报, 2020, 41(12): 49-54. DOI: 10.12073/j.hjxb.20200815002
LIU Tianyuan, BAO Jinsong, WANG Junliang, ZHENG Xiaohu, WANG Jiacheng. Adaptive edge detection of molten pool based on coarse-grained regularization in restricted solution space[J]. TRANSACTIONS OF THE CHINA WELDING INSTITUTION, 2020, 41(12): 49-54. DOI: 10.12073/j.hjxb.20200815002
Citation: LIU Tianyuan, BAO Jinsong, WANG Junliang, ZHENG Xiaohu, WANG Jiacheng. Adaptive edge detection of molten pool based on coarse-grained regularization in restricted solution space[J]. TRANSACTIONS OF THE CHINA WELDING INSTITUTION, 2020, 41(12): 49-54. DOI: 10.12073/j.hjxb.20200815002

受限解空间下粗粒度正则化的熔池边缘自适应检测方法

Adaptive edge detection of molten pool based on coarse-grained regularization in restricted solution space

  • 摘要: 边缘检测是熔池图像处理的关键步骤. 鉴于熔池区域弧光变化剧烈,依赖人为设置阈值的边缘检测方法难以适应弧光的动态变化过程,文中提出了一种基于深度学习的熔池边缘提取模式. 首先对原始熔池图像进行像素级标注和数据增广以构建数据集;其次结合先验知识提出了一种受限解空间下的粗粒度正则化方法(coarse grained regularization method in restricted solution space, CGRRSS)以增强边缘特征;最后从定量和定性两方面将所提方法与传统方法进行了对比. 结果表明,所提方法对边缘点的召回率较高,所获熔池边缘更连续且对伪边缘具有更好的抑制作用. 单幅图像检测时间6.2 ms,可满足在线监测需求.

     

    Abstract: Edge detection is a key step in the image processing of the molten pool. In view of the sharp changes in the arc in the molten pool area, the edge detection method that relies on artificially setting the threshold is difficult to adapt to the dynamic change of the arc. This paper proposes a deep learning-based edge extraction mode of the molten pool. Firstly, pixel-level annotation and data augmentation are performed on the original molten pool image to build a dataset. Secondly, a coarse-grained regularization method in restricted solution space (CGRRSS) is proposed to enhance edge features. Finally, the proposed method is compared with the traditional methods in both quantitative and qualitative aspects. The results show that the proposed method has a higher recall of edge points, the obtained molten pool edge is more continuous and has a better suppression effect on false edges. The detection time of a single image is 6.2 ms, which can meet the needs of online monitoring.

     

/

返回文章
返回