基于改进PSPNet的手机LCD屏幕表面缺陷检测

Surface Defect Detection of Mobile Phone LCD Screen Based on Improved PSPNet

摘要:

手机屏幕是智能手机的关键部件，其品质优劣直接影响到用户的使用体验；因此，手机屏幕缺陷检测成为工业生产中的重要环节；然而，手机LCD屏幕的表面缺陷检测目前还存在检测精确度低、模型参数较多等问题，无法满足实际工业生产需求；为了解决这些问题，对现有的缺陷检测算法和经典语义分割模型进行了研究，提出一种基于改进PSPNet的手机LCD屏幕表面缺陷检测模型；模型采用MobileNetV3作为特征提取网络，有效减少了模型参数；采用多尺度金字塔池化模块，进一步整合多尺度上下文信息，提高了模型的特征提取能力，有效应对屏幕图像中缺陷尺寸微小、边界模糊、相同缺陷尺寸差异较大的问题；同时，通过引入注意力机制，增强了模型的鲁棒性；实验结果表明，在SQ、Mura、TP、Line四种类型的手机LCD屏幕表面缺陷检测上，改进后的模型准确度明显优于基线模型。

Abstract:

As the core component of a phone, the quality of screen is directly related to the user's experience. Therefore, mobile phone screen defect detection has become an important part of industrial production. However, the surface defect detection of mobile LCD screens still faces problems such as low detection accuracy and a large number of model parameters, which cannot meet the actual industrial production needs. After studing existing defect detection algorithms and classical semantic segmentation models, an improved mobile phone LCD screen defect detection model based on PSPNet is proposed to solve the problems. MobileNetV3 is used to replace the original ResNet50 as the backbone, which effectively reduces the model parameters and shortens the training time. A multi-scale pyramid pooling module is proposed to effectively integrate multi-scale contextual information, which improves the feature extraction ability of the model. It also effectively addresses the issues of small defect sizes, blurred boundaries, and significant differences in the size of the same defect in screen images Meanwhile, the introduction of attention mechanism improves the anti-interference ability of the model. The experimental results show that the accuracy of the improved model on the mobile phone LCD screen dataset has significantly better accuracy than other traditional semantic segmentation models.