大数据驱动的非球面光学表面缺陷检测研究  

Research on non spherical optical surface defect detection driven by big data

作　　者：彭雪梅[1] 黄建军[1] PENG Xuemei;HUANG Jianjun(School of Computer Information Engineering Nanchang Institute of Technology,Nanchang 330044,China)

机构地区：[1]南昌理工学院计算机信息工程学院,南昌330044

出　　处：《激光杂志》2025年第4期252-256,共5页Laser Journal

基　　金：江西省教育厅科学技术项目(No.GJJ2202717)。

摘　　要：光学技术的快速发展使非球面光学元件在成像、通信等领域的应用日益广泛,其表面缺陷检测成为确保产品质量的关键环节。为满足更高精度的检测需求,设计了大数据驱动的非球面光学表面缺陷检测方法。在大数据驱动下,设计非球面光学表面图像采集装置,由显微光学成像系统、分光棱镜、光纤照明、机械调整台、电动转台构成,实施非球面光学表面图像的大数据采集。在大数据驱动下,采用小波阈值去噪方法对采集的非球面光学表面缺陷图像实施去噪处理。选定YOLOv3作为基础架构,对三个方向实施针对性改进,通过改进后的YOLOv3模型实现非球面光学表面缺陷检测。测试结果表明,所设计的方法对于5种实验元件的表面缺陷平均尺寸测量偏差较低,尤其是对于抛光非球面棱镜的表面缺陷,其平均尺寸测量偏差最低。此外,该方法对于五种实验元件的表面伪缺陷响应系数较低,意味着它不容易受到伪缺陷的影响。The rapid development of optical technology has made the application of aspherical optical components in imaging,communication and other fields increasingly widespread,and surface defect detection has become a key link to ensure product quality.To meet the demand for higher precision detection,this paper designs a big data-driven method for detecting defects on non spherical optical surfaces.Under the drive of big data,design an aspherical optical surface image acquisition device,consisting of a microscopic optical imaging system,a splitter prism,fiber optic lighting,a mechanical adjustment table,and an electric turntable,to implement big data acquisition of aspherical optical surface images.Under the drive of big data,the wavelet threshold denoising method is used to denoise the collected images of aspherical optical surface defects.Select YOLOv3 as the infrastructure,implement targeted improvements in three directions,and achieve non spherical optical surface defect detection through the improved YOLOv3 model.The test results indicate that the designed method has a lower average measurement deviation for the surface defects of the five experimental components,especially for the surface defects of polished aspherical prisms,which have the lowest average measurement deviation.In addition,this method has a lower response coefficient to surface pseudo defects for the five experimental components,which means it is less susceptible to the influence of pseudo defects.

关 键 词：大数据技术 非球面光学元件 表面缺陷检测 改进YOLOv3模型 CBAM混合注意力机制 

分 类 号：TN911[电子电信—通信与信息系统]