基于结构光视觉的偏光片内部压痕缺陷检测方法研究

The internal transparent dent defects of polarizer can hardly imaging under common illumination. The theoretical basis and design method derived from the studies of the mechanics of imaging enhancement and the depth measurement would be instructive to the further improvement of inspection accuracy and speed. . This project will investigate the models of polarizer, dent defect, structured light and camera, and followed by the propagation of the homogeneous light and the structured light in polarizer according to the ray trajectory law. The intensity distribution of reflective and transmitted images will then be calculated to represent how the effect of defect imaging enhancement depends on the parameters of the imaging system, so the mechanics of defect imaging enhancement is established. The project will also work on, according to the mechanics, how to design and produce the structured light, study, and how to quickly imaging a piece of large polarizer with the active scan approach. A straight and fast image processing algorithm will be designed which can work well with the structured light illumination and the active scan imaging. The measurement approach of dent depth will be studied as well as the automatic identifying and classifying methods of the dent defect based on ANN. . This project proposes a new approach in that the internal transparent dent defects of polarizer are inspected under the structured light illumination, and it also reveals the mechanics of defect imaging enhancement. So this project would have important theory significance and application potential to the internal defects inspection of polarizer and other multilayer films products.

在普通照明条件下，偏光片内部透明的压痕缺陷难以成像检测，研究该缺陷光学模型与成像增强机理，为进一步提高缺陷检测准确率和检测速度，提供方法和理论依据。. 研究偏光片、压痕缺陷、结构光源和相机的光学模型，然后基于光线轨迹法，研究均匀光与结构光在缺陷样品中的传播规律，计算出反射和透射图像的强度分布，明确缺陷图像增强效果与成像系统参数之间的关系，从而阐明缺陷图像增强的机理；根据增强机理，研究结构光源设计及其产生方法，进而研究快速覆盖整幅大尺寸偏光片的主动扫描方法；研究与结构光照明、主动成像方法相匹配的快速图像处理与缺陷识别算法；研究压痕深度测量方法，建立基于人工神经网络的压痕缺陷自动识别与分类方法。. 本项目首次提出采用结构光照明检测偏光片内部透明压痕缺陷的方法，并揭示成像增强机理、缺陷深度与产品分类之间的关系模型，对偏光片及其它多层透明薄膜的缺陷检测，具有重要的理论意义和应用价值。

针对偏光片内部难以成像检测的压痕缺陷，本项目采用二元条纹结构光照明，极大增强了缺陷成像对比度，并将该类缺陷建模为平凸微透镜，在TracePro里构建了缺陷成像仿真系统，仿真并实验研究了缺陷成像效果与检测系统参数之间的关系。通过仿真与实验研究，给出了结构光成像增强的机理：二元线型结构光相当于引入了一个一维光阑，增强了缺陷（平凸透镜）的成像作用，从而提高了缺陷对比度。针对极细微的压痕缺陷，本项目基于上述结构光检测的研究基础，进一步提出了“边缘光”和“饱和成像”两种检测新方法，能有效提高缺陷检测准确率。基于微透镜成像和相机的饱和效应，给出了边缘光检测和饱和成像检测的机理。针对饱和成像检测方法，提出一种具有平移、尺度以及旋转不变性的条纹结构光成像的饱和度定义——建立一个对比度vs饱和度的曲线簇经验模型，利用五个饱和度位置上的对比度测量数据，即可通过参数优化方法寻找到最佳饱和度。提出了一种基于饱和成像和回归估计的压痕三维形态测量新方法，采集压痕饱和成像数据，与共聚焦显微镜测量数据建立标记的数据集，在该集上进行降维，在低维的空间使用支持向量回归以建立饱和成像与缺陷形态参数的关系，把这种回归关系应用于新的缺陷饱和成像样本即可估测缺陷形态。缺陷深度平均相对误差小于3.6%，缺陷宽度平均相对误差小于1.7%，计算缺陷形态的时间小于0.01s。这一估测方法可以在自动检测外观缺陷的同时，完成压痕缺陷形态测量，可以极大提高质检效率。本项目还开展了RPCA算法（APG/IALM）、空间纹理滤波算法、基于机器学习的缺陷样本生成和分类算法等方面的研究。