文献类型：期刊文献

中文题名：缺陷检测中的标签噪声形态恢复方法及其应用

英文题名：Label Noise Morphology Recovery Method for Defect Detection and Its Application

作者：张富照[1];苏林[1];陈元昊[1];易建军[1];盛涛[2];郑金华[2]

机构：[1]华东理工大学机械与动力工程学院,上海200237;[2]上海复合材料科技有限公司,上海201112

年份：2023

卷号：40

期号：6

起止页码：159

中文期刊名：上海航天（中英文）

外文期刊名：Aerospace Shanghai（Chinese&English）

收录：CSTPCD;;CSCD:【CSCD2023_2024】；

语种：中文

中文关键词：缺陷检测;噪声过滤;深度学习;碳纤维复合材料

外文关键词：defect detection;noise filtering;deep learning;carbon fiber composite

摘要：为了保证碳纤维材料产品的可靠性,消除各种可能存在的缺陷,有必要采取有效的手段对其质量进行检查。结合图像识别算法的基于X射线无损检测技术被认为是一种快速有效的解决方案。然而,加工材料的表面通常附有包含各种信息的标签,这些标签会在检测中对缺陷的识别造成干扰,甚至被误检为缺陷。主要研究基于图像特征的产品标签噪声恢复方法及其在缺陷检测中的应用,该方法可以有效地消除噪声,而不影响其余的图像信息,从而确保算法正确识别材料中的缺陷。
In order to ensure the reliability of carbon fiber material products and eliminate various possible defects,it is necessary to adopt effective means to inspect their quality.The X-ray based non-destructive testing technique combined with image recognition algorithms is considered to be a fast and effective solution.However,the surface of the processed material is usually accompanied by labels containing various information,which will interfere with the identification of defects during inspection or even result in false detection.This paper focuses on an image feature-based noise recovery method for product labels and its application in defect detection.The method can effectively eliminate noise without affecting the rest of the image information,thus ensuring that the algorithm can correctly identify the defects in the material.