Page 51 - 无损检测2024年第六期
P. 51
缺陷的智能识别与分类专题
缺陷的智能识别与分类专题
DOI:10.11973/wsjc202406004
基于深度学习的焊缝缺陷 X 射线检测图像
识别与增强
王树森 ,李 萍 ,黄大伟 ,李晓庆 ,吴中华 ,张忠仁 ,王 爽 ,田 双 ,杨毅德 2
2
1
1
2
2
3
2
2
(1. 大连理工大学 材料科学与工程学院,大连 116000;2. 大连船舶重工集团有限公司 质量检验部,大连 116000;
3. 大连市锅炉压力容器检验研究院 承压检验部,大连 116000)
摘 要:为了提高焊缝缺陷X射线图像识别的准确率,需要采用有效的图像增强技术,笔者研
究了不同图像增强方法对焊缝图像质量的影响,用峰值信噪比、结构相似度、结构清晰度、信息熵
等参数对图像增强质量进行评价。试验结果表明,直方图均衡化(HE)与限制对比度自适应直方图
均衡化(CLAHE)有较好的对比度增强效果,非局部均值滤波(NLM)与小波降噪(DWT)的去噪综
合表现较好。基于CLAHE-NLM的图像增强处理可以更有效地帮助深度学习模型进行焊缝缺陷
分类识别,焊缝缺陷分类的准确率与F 1 值达97. 6%和96. 93%,相较于未增强处理的数据集提高了
3. 2%与5. 23%。
关键词:图像增强;深度学习;焊缝缺陷;X射线
中图分类号:TP391;TG115.28 文献标志码:A 文章编号:1000-6656(2024)06-0017-07
Image recognition and enhancement of X-ray detection of weld defects based on deep learning
WANG Shusen , LI Ping , HUANG Dawei , LI Xiaoqing , WU Zhonghua , ZHANG Zhongren ,
1
1
2
2
2
2
WANG Shuang , TIAN Shuang , YANG Yide 2
3
2
(1. College of Material Science and Engineering, Dalian University of Technology, Dalian 116000, China;
2. Department of Quality Inspection, Dalian Shipbuilding Industry Co., Ltd., Dalian 116000, China;
3. Department of Pressure Inspection, Dalian Boiler and Pressure Vessel Inspection Institute, Dalian 116000, China)
Abstract: In order to enhance the accuracy of X-ray image recognition for weld seam defects, it is essential to employ
effective image enhancement techniques. this study investigates investigated the impact of different image enhancement
techniques on the quality of weld seam images. The parametersEvaluation metrics such as Peak Signal-to-Noise Ratio
(PSNR), Structural Similarity Index (SSIM), Structural Clarity, and Information Entropy wereare employed to assess
the quality of image enhancement. Experimental results indicated that histogram equalization (HE) and contrast-limited
adaptive histogram equalization (CLAHE) exhibited superior contrast enhancement effects, while non-local means filtering
(NLM) and discrete wavelet transform (DWT) performed well in noise reduction. Image enhancement processing based
on CLAHE-NLM proveds to be more effective in assisting deep learning models for weld seam defect classification and
recognition. The accuracy and F score of weld seam defect classification reached 97.6% and 96.93%, respectively,
1
representing an improvement of 3.2% and 5.23% compared to the data set without enhancement preprocessing.
Key words: image enhancement; deep learning; weld defect; X-ray
X射线检测是常用的焊缝缺陷检测技术,但缺
收稿日期:2024-01-18
基金项目:国家自然科学基金项目(51901035) 陷的形态、大小、方向具有不确定性,使得射线底片
作者简介:王树森(1998—), 男, 硕士研究生, 主要研究方向为 [1]
评价较为复杂 ,仅依赖人工进行焊缝质量检测,
无损检测、深度学习
时间较长且效率低,无法满足现代化焊接技术的要
通信作者:李 萍(1969—), 女, 博士, 副教授, 硕士生导师,主
要研究方向为材料无损表征与评价、模式识别 求。因此,将深度学习应用于焊缝缺陷的智能化检
17
2024 年 第 46 卷 第 6 期
无损检测

