移动电话
生成语法
计算机科学
对抗制
曲面(拓扑)
电话
人工智能
电信
数学
几何学
语言学
哲学
作者
Zhuojia Ma,Meiqin Liu,Senlin Zhang,Shanling Dong,Ping Wei
标识
DOI:10.1109/ccdc62350.2024.10588188
摘要
This paper proposes a novel defect detection method for mobile phone screens based on Generative Adversarial Networks (GANs). It aims at solving the overfitting problem due to insufficient data and enhancing the effectiveness of the detection network. To achieve the goal, we first present Multiscale Generative Adversarial Networks (MsGANs) to expand the dataset. As two important components in MsGANs, multi-scale generators and pyramidal discriminators are obtained by optimizing the structure of Conditional Generative Adversarial Networks (CGANs). Multi-scale generators are built from coarse to fine, where the global generator produces images with lower resolution, and the local enhancer provides missing details to ensure the creation of excellent images. Pyramidal discriminators utilize three distinct scales of image feature pyramids. Further-more, we establish the multi-label classification network to tackle the defect detection problem that multiple categories of defects are involved in a single sample. A pre-trained ResNet-50 is used to extract high-level image features. Finally, the experimental results demonstrate the effectiveness of MsGANs combined with ResNet-50 for mobile phone screen defect detection.
科研通智能强力驱动
Strongly Powered by AbleSci AI