盖板玻璃外观检测的高光谱线扫成像方法

doi:10.3964/j.issn.1000-0593(2025)03-0616-07

摘要
参考文献
相关文章 (15)

全文: PDF (2889 KB)
输出: BibTeX | EndNote (RIS)

摘要：玻璃盖板是智能终端产品的重要组成配件，表面光滑且透明，其外观检测是光学成像检测领域的重要难题之一。目前，常规检测方法主要基于可见光影像，但是由于纹理相似性，常将污渍误判为瑕疵，导致将良品被判定为次品，从而增加工业成本。为了克服上述问题，提出基于高光谱技术的盖板玻璃污渍瑕疵检测方法。该方法通过对采集的高光谱数据进行最优特征光谱选择和建立定量检测模型，挑选出关键特征波段并实现对污渍瑕疵的精确检测。利用油墨区和AA透明区的光学特性，采用线性光源透射的成像方式，通过专业高光谱线扫设备，有效采集50个手机盖板玻璃高光谱影像，制作了包括无污无瑕盖板样本500个，及玻璃指纹、胶质物质、清洁剂、灰尘等4种污渍和划痕瑕疵样本各100个污渍瑕疵的数据集。基于上述高光谱影像数据集，构建了一种综合考虑污渍瑕疵的光谱特性和各个特征波段的贡献率与重要性的波段选择方法，精选出可有效区分污渍与瑕疵的8个特征波段（502、526、567、689、711、789、818和888 nm）。利用机器学习算法进行检测，实验结果表明，8个精选高光谱波段对区分污渍瑕疵具有良好的效果，准确率达95.4%，误分率仅为4.7%。高光谱影像能捕捉瑕疵和污渍在光谱上的差异，实现更精确地检测，为手机盖板玻璃质量检测提供了一种可行的新方法，可为实际应用中设计低成本高光谱瑕疵污渍检测相机提供参考。

关键词：高光谱技术；污渍瑕疵检测；特征波段选择；随机森林算法

Abstract：Cover glass is an important component of intelligent terminal products, with a smooth, transparent surface and complex and variable characteristics. Its appearance inspection is one of the important challenges in optical imaging detection. Currently, conventional detection methods are mainly based on visible light images. Still, due to texture similarity, stains are often misjudged as defects, making good products judged as defective, thereby increasing industrial costs. To overcome the above problems, this article proposes a method for detecting stains and defects on cover glass based on hyperspectral technology. This method selects the hyperspectral data's optimal feature spectrum and establishes a quantitative detection model to select key feature bands and accurately detect stains and defects. This article utilizes the optical properties of the ink and AA transparent areas and adopts a linear light source transmission imaging method. Through professional hyperspectral line scanning equipment, 50 hyperspectral images of mobile phone cover glass were effectively collected. A stain defect dataset was created, including 500 samples of clean and flawless cover plates and 100 samples of four types of stains and scratch defects, including glass fingerprints, adhesive substances, cleaning agents, and dust. Based on the above hyperspectral image dataset, this paper constructs a band selection method that comprehensively considers the spectral characteristics of stains and defects and the contribution and importance of each feature band. Eight feature bands (502, 526, 567, 689, 711, 789, 818, and 888 nm) that can effectively distinguish stains and defects are selected. Using machine learning algorithms for detection, experimental results show that 8 selected hyperspectral bands perform well in distinguishing stains and defects, with an accuracy rate of 95.4% and an error rate of only 4.7%. Hyperspectral imaging can capture the differences between defects and stains in the spectrum, achieving more accurate detection and providing a feasible new method for quality inspection of mobile phone cover glass. It can provide a reference for designing low-cost hyperspectral defect and stain detection cameras in practical applications.

Key words：Hyperspectral; Stain defect detection; Feature band selection; Random forest

收稿日期: 2023-11-14 修订日期: 2024-07-03

中图分类号:

O433.4

基金资助: 河南省科学院科研启动经费项目（242025005），国家重大科研仪器研制项目（11727806）资助

通讯作者: 巫兆聪 E-mail: zcwoo@whu.edu.cn

作者简介: 沈冠廷，1999年生，武汉大学遥感信息工程学院硕士研究生 e-mail: 2017312590088@whu.edu.cn

引用本文:

沈冠廷，饶可奕，方瑞欣，张学敏，巫兆聪. 盖板玻璃外观检测的高光谱线扫成像方法[J]. 光谱学与光谱分析, 2025, 45(03): 616-622.
SHEN Guan-ting, RAO Ke-yi, FANG Rui-xin, ZHANG Xue-min, WU Zhao-cong. A Method for Classifying Stains and Defects on Mobile Phone Cover Glass Based on Hyperspectral Line Scan Imaging. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(03): 616-622.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2025)03-0616-07 或 https://www.gpxygpfx.com/CN/Y2025/V45/I03/616

[1] CUI Yan，PENG Ke，YANG Yu-e，et al（崔焱，彭可，杨玉娥，等）. Manufacturing Automation（制造业自动化），2023，45（7）：75.
[2] HE Yi-lin，TAO Xiao（贺旖琳，陶肖）. Industrial Control Computer（工业控制计算机），2022，35（12）：74.
[3] Latham V H, Nixon M S, Mayers C. Glass Technology, 1986，27：188.
[4] Li C，Zhang X，Huang Y，et al. Computers & Industrial Engineering, 2020，146：106530.
[5] ZHANG Jian-guo，LI Ying，QI Jia-kun，et al（张建国，李颖，齐家坤，等）. Journal of Applied Optics（应用光学），2020，41(5)：984.
[6] Jian C X, Gao J, Ao Y H. Applied Soft Computing，2017，52：348.
[7] LUO Gen，NI Jun（罗根，倪军）. Journal of Electronic Measurement and Instrumentation（电子测量与仪器学报），2018，32（2）：92.
[8] Takhar G，Prakash C，Mittal N，et al. Comparative Analysis of Background Subtraction Techniques and Application, International Conference on Recent Advances & Innovations in Engineering IEEE, 2017.
[9] Zhang Z, Xu Y，Yang J，et al. IEEE Access, 2015, 3: 490.
[10] Li D, Liang L Q, Zhang W J，et al. The International Journal of Advanced Manufacturing Technology，2014，73（9）：1605.
[11] Tsai D M，Tseng Y H，Fan S K M. Defect Inspection of Liquid-Crystal-Display (LCD) Panels in Repetilive Pattern Images Using 2D Fourier Image Reconstruction, International Conference on Autonomic and Autonomous Systems，2018.
[12] LI Wei-chao，CHEN Zhi-hao，ZHANG Xie，et al（李伟朝，陈志豪，张勰，等）. Computer Measurement & Control（计算机测量与控制），2023，31（7）：99.
[13] WU Ji-gang，CHENG Yuan，SHAO Jun，et al（伍济钢，成远，邵俊，等）. Chinese Journal of Liquid Crystals and Displays（液晶与显示），2021，36（12）：1728.
[14] WU Chuang，YU Da-yong（吴闯，于大泳）. Software Engineering（软件工程），2021，24（12）：6.