Metal Surface Defect Detection Method Based on Improved Cascade R-CNN

IF 2.6 3区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Journal of Computing and Information Science in Engineering Pub Date : 2023-10-20 DOI:10.1115/1.4063860

Yani Wang, Xiang Wang, Ruiyang Hao, Bingyu Lu, Biqing Huang

{"title":"Metal Surface Defect Detection Method Based on Improved Cascade R-CNN","authors":"Yani Wang, Xiang Wang, Ruiyang Hao, Bingyu Lu, Biqing Huang","doi":"10.1115/1.4063860","DOIUrl":null,"url":null,"abstract":"Abstract In contemporary industrial systems, ensuring the quality of object surfaces has become an essential and inescapable aspect of factory inspections. Cascade Regional Convolutional Neural Network (Cascade R-CNN), an object detection and instance segmentation algorithm based on deep learning, has been widely applied in numerous industrial applications. Nonetheless, there is still space for improving the detection of defects on metal surfaces. This paper proposes an enhanced metal defect detection method based on Cascade R-CNN. Specifically, the improved backbone network is employed to acquire the features of images, which enables more precise localization. Additionally, up and down sampling is combined to extract multi-scale defect feature maps, and contrast histogram equalization enhancement is utilized to tackle the issue of unclear contrast in the data. Experimental results demonstrate that the proposed approach achieves a mean Average Precision (mAP) of 0.754 on the NEU-DET dataset, and outperforms the Cascade R-CNN model by 9.2%.","PeriodicalId":54856,"journal":{"name":"Journal of Computing and Information Science in Engineering","volume":"11 1","pages":"0"},"PeriodicalIF":2.6000,"publicationDate":"2023-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Computing and Information Science in Engineering","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1115/1.4063860","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}

引用次数: 0

Abstract

Abstract In contemporary industrial systems, ensuring the quality of object surfaces has become an essential and inescapable aspect of factory inspections. Cascade Regional Convolutional Neural Network (Cascade R-CNN), an object detection and instance segmentation algorithm based on deep learning, has been widely applied in numerous industrial applications. Nonetheless, there is still space for improving the detection of defects on metal surfaces. This paper proposes an enhanced metal defect detection method based on Cascade R-CNN. Specifically, the improved backbone network is employed to acquire the features of images, which enables more precise localization. Additionally, up and down sampling is combined to extract multi-scale defect feature maps, and contrast histogram equalization enhancement is utilized to tackle the issue of unclear contrast in the data. Experimental results demonstrate that the proposed approach achieves a mean Average Precision (mAP) of 0.754 on the NEU-DET dataset, and outperforms the Cascade R-CNN model by 9.2%.

查看原文本刊更多论文

基于改进级联R-CNN的金属表面缺陷检测方法

在当代工业系统中，保证物体表面的质量已经成为工厂检查必不可少的和不可避免的方面。Cascade区域卷积神经网络(Cascade R-CNN)是一种基于深度学习的目标检测和实例分割算法，在众多工业应用中得到了广泛的应用。尽管如此，金属表面缺陷的检测仍有改进的空间。本文提出了一种基于级联R-CNN的增强金属缺陷检测方法。具体来说，利用改进的骨干网络获取图像的特征，使定位更加精确。并结合上下采样提取多尺度缺陷特征图，利用对比度直方图均衡化增强解决数据对比度不清的问题。实验结果表明，该方法在nue - det数据集上的平均平均精度(mAP)为0.754，比Cascade R-CNN模型高出9.2%。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of Computing and Information Science in Engineering 工程技术-工程：制造

CiteScore

6.30

自引率

12.90%

发文量

100

审稿时长

6 months

期刊介绍： The ASME Journal of Computing and Information Science in Engineering (JCISE) publishes articles related to Algorithms, Computational Methods, Computing Infrastructure, Computer-Interpretable Representations, Human-Computer Interfaces, Information Science, and/or System Architectures that aim to improve some aspect of product and system lifecycle (e.g., design, manufacturing, operation, maintenance, disposal, recycling etc.). Applications considered in JCISE manuscripts should be relevant to the mechanical engineering discipline. Papers can be focused on fundamental research leading to new methods, or adaptation of existing methods for new applications. Scope: Advanced Computing Infrastructure; Artificial Intelligence; Big Data and Analytics; Collaborative Design; Computer Aided Design; Computer Aided Engineering; Computer Aided Manufacturing; Computational Foundations for Additive Manufacturing; Computational Foundations for Engineering Optimization; Computational Geometry; Computational Metrology; Computational Synthesis; Conceptual Design; Cybermanufacturing; Cyber Physical Security for Factories; Cyber Physical System Design and Operation; Data-Driven Engineering Applications; Engineering Informatics; Geometric Reasoning; GPU Computing for Design and Manufacturing; Human Computer Interfaces/Interactions; Industrial Internet of Things; Knowledge Engineering; Information Management; Inverse Methods for Engineering Applications; Machine Learning for Engineering Applications; Manufacturing Planning; Manufacturing Automation; Model-based Systems Engineering; Multiphysics Modeling and Simulation; Multiscale Modeling and Simulation; Multidisciplinary Optimization; Physics-Based Simulations; Process Modeling for Engineering Applications; Qualification, Verification and Validation of Computational Models; Symbolic Computing for Engineering Applications; Tolerance Modeling; Topology and Shape Optimization; Virtual and Augmented Reality Environments; Virtual Prototyping