复合材料科学与工程 ›› 2024, Vol. 0 ›› Issue (1): 83-88.DOI: 10.19936/j.cnki.2096-8000.20240128.011

• 应用研究 • 上一篇    下一篇

基于Mask R-CNN的复合材料夹杂缺陷自动检测研究

李磊磊1, 王明泉1*, 赵付宝2, 朱焕宇1, 丰晓钰1, 谢绍鹏1   

  1. 1.中北大学 仪器科学与动态测试教育部重点实验室,太原 030051;
    2.山东非金属材料研究所,济南 250000
  • 收稿日期:2023-10-19 出版日期:2024-01-28 发布日期:2024-02-27
  • 通讯作者: 王明泉(1970—),男,博士,教授,硕士生导师,研究方向为图像处理与识别、成像技术,wangmq@nuc.edu.cn。
  • 作者简介:李磊磊(1999—),男,硕士研究生,研究方向为复合材料缺陷检测与深度学习。
  • 基金资助:
    山西省高等学校科技创新项目(2020L0624)

Research on automatic detection of composite inclusion defects based on Mask R-CNN

LI Leilei1, WANG Mingquan1*, ZHAO Fubao2, ZHU Huanyu1, FENG Xiaoyu1, XIE Shaopeng1   

  1. 1. MOE Key Laboratory of Instrumentation Science and Dynamic Measurement, North University of China, Taiyuan 030051, China;
    2. Shandong Institute of Nonmetallic Materials, Jinan 250000, China
  • Received:2023-10-19 Online:2024-01-28 Published:2024-02-27

摘要: 为提高复合材料夹杂缺陷的检测效率,本文提出利用深度学习网络设计一种夹杂缺陷自动检测系统。在图像预处理环节采用两级反锐化掩膜算法突出夹杂缺陷特征,构建复合材料夹杂缺陷图像数据库;采用Mask R-CNN网络模型,经过网络模型训练,得到最优权重参数,最终设计实现缺陷检测软件系统。实验结果表明,Mask R-CNN算法网络准确率达94.6%,召回率达92.4%,AP值达87.3%。该系统应用方便快捷,将有效提高一线人员的缺陷检测效率和检测精度。

关键词: 反锐化掩膜, 图像处理, 深度学习, 缺陷检测, 系统设计, 复合材料

Abstract: In order to improve the detection efficiency of composite inclusion defects, an automatic inclusion defect detection system based on deep learning network is proposed in this paper. In the process of image preprocessing, a two-stage unsharping mask algorithm is used to highlight the features of inclusion defects, and a composite image database of inclusion defects is constructed. The Mask R-CNN network model is used, and the optimal weight parameters are obtained through network model training. Finally, the defect detection software system is designed and realized. The experimental results show that the network accuracy of Mask R-CNN algorithm is 94.6%, the recall rate is 92.4%, and the AP value is 87.3%. The system is convenient and fast in application, and will effectively improve the efficiency and accuracy of defect detection for front-line personnel.

Key words: unsharp mask, image processing, deep learning, defect detection, system design, composites

中图分类号: