基于U-P-Net的手机玻璃屏幕缺陷分割 |
| |
引用本文: | 李墨,陈志豪,吴宗泽.基于U-P-Net的手机玻璃屏幕缺陷分割[J].计算机测量与控制,2023,31(8):231-237. |
| |
作者姓名: | 李墨 陈志豪 吴宗泽 |
| |
作者单位: | 广东工业大学大学 计算机学院,,广东工业大学大学 自动化学院 |
| |
基金项目: | 国家重点研发计划项目(2020AAA0108304),国家自然科学(62073088,U1911401)。 |
| |
摘 要: | 随着科技的发展及电子设备的普及,玻璃屏幕质量成为电子设备和其他产品的重要考虑因素;而玻璃外观缺陷检测是玻璃质量检测中最重要的环节,这也是保证产出高品质、高性能的玻璃产品的关键环节;目前玻璃表面缺陷检测方法存在无目标训练图像资源消耗、检测精度较低、复杂特征信息难以提取等问题;因此,为了解决上述问题,提出了一种基于U-pyramid pooling module-Net(U-P-Net)的手机玻璃屏幕缺陷分割模型;采用超像素预处理,有效地降低了原始图像的复杂度;采用ResNet50作为分类网络,减少无目标训练图像造成的资源浪费,提高训练效率;U-P-Net被提出,有效地聚合了不同区域的上下文信息,提高了获取全局信息的能力;实验结果表明,所设计的基于U-P-Net玻璃缺陷分割算法分割精度明显优于其它传统卷积神经网络分割方法,证明了该框架在移动屏幕数据集上的有效性.
|
关 键 词: | 缺陷检测 超像素预处理 ResNet50 U-P-Net 金字塔池化 手机玻璃屏幕 |
收稿时间: | 2023/3/2 0:00:00 |
修稿时间: | 2023/3/14 0:00:00 |
Defect segmentation of mobile phone screen based on U-P-Net |
| |
Abstract: | With the development of science and technology and the popularity of electronic equipment, glass screen quality has become an important consideration for electronic equipment and other products; The detection of glass appearance defects is the most important link in glass quality detection, which is also the key link to ensure the production of high quality and high-performance glass products; At present, there are some problems in the detection methods of glass surface defects, such as resource consumption of target-free training image, low detection accuracy and difficult extraction of complex feature information. Therefore, to solve the above problems, a defect segmentation model of mobile phone glass screens based on U-pyramid pooling module-Net(U-P-Net) is proposed. Superpixel preprocessing is used to reduce the complexity of the original image effectively. ResNet50 was used as a classification network to reduce the waste of resources caused by training images without targets and improve training efficiency. U-P-Net is proposed, which aggregates the context information of different regions effectively and improves the ability to obtain global information. Experimental results show that the proposed U-P-Net glass defect segmentation algorithm is significantly superior to other traditional convolutional neural network segmentation methods, which proves the effectiveness of the framework on mobile screen data sets. |
| |
Keywords: | defect detection superpixel pre-processing ResNet50 pyramid pooling mobile phone glass screen |
|
| 点击此处可从《计算机测量与控制》浏览原始摘要信息 |
|
点击此处可从《计算机测量与控制》下载全文 |
|