摘要
由于加工工艺的影响,应用零件表面划痕形态也会呈现出多种不同的表现形式,受到背景纹理样式的干扰,零件表面的细微划痕依然难以得到准确分辨与检测,为解决此问题,设计基于深度学习的零件表面细微划痕检测方法。在深度学习网络的彩色空间环境中,对滞后多阈值进行分割处理,再联合投资回报率(return on investment, ROI)提取权限条件,完成基于深度学习的划痕特征参量提取。在此基础上,建立边缘模板,通过处置细微划痕配准需求的方式,实现对划痕的准确定位,完成基于深度学习零件表面细微划痕检测方法的设计与应用。与机器视觉型检测方法相比,深度学习型检测方法能够根据划痕所表现出的具体形态,对其进行检测与分辨,可避免背景纹理样式对于零件加工工艺的影响。
Due to the influence of processing technology,application parts surface scratch form will also present a variety of different forms,disturbed by the background texture style,the parts surface fine scratch is still difficult to get accurate resolution and detection,to solve this problem,design based on deep learning parts surface fine scratch detection method.In the color spatial environment of the deep learning network,the lag multi-threshold is divided,and then the return on investment(ROI)is extracted to complete the scratch feature parameter extraction based on deep learning.On this basis,the edge template is established to realize the accurate positioning of scratches by handling the registration requirements of fine scratches,and complete the design and application of the fine scratch detection method of the surface based on deep learning parts.Compared with machine vision detection method,deep learning detection method can detect and distinguish the scratches according to the specific form shown by the scratch,which can avoid the influence of background texture style on the processing process of parts.
作者
张凯
ZHANG Kai(Shandong University of Engineering and Vocational Technology,Jinan 250200,Shandong,China)
出处
《流体测量与控制》
2023年第5期11-14,共4页
Fluid Measurement & Control
基金
中国民办教育协会2022年度规划课题(CANFZG22354)。
关键词
深度学习
细微划痕
彩色空间
滞后多阈值
ROI权限
边缘模板
deep learning
fine scratch
color space
lag multi-threshold
return on investment(ROI)permission
edge template
