Attention-relation network for mobile phone screen defect classification via a few samples

How to use a few defect samples to complete the defect classification is a key challenge in the production of mobile phone screens.An attention-relation network for the mobile phone screen defect classification is proposed in this paper.The architecture of the attention-relation network contains two modules:a feature extract module and a feature metric module.Different from other few-shot models,an attention mechanism is applied to metric learning in our model to measure the distance between features,so as to pay attention to the correlation between features and suppress unwanted information.Besides,we combine dilated convolution and skip connection to extract more feature information for follow-up processing.We validate attention-relation network on the mobile phone screen defect dataset.The experimental results show that the classification accuracy of the attentionrelation network is 0.9486 under the 5-way 1-shot training strategy and 0.9039 under the 5-way 5-shot setting.It achieves the excellent effect of classification for mobile phone screen defects and outperforms with dominant advantages.

A graph-based two-stage classification network for mobile screen defect inspection

Defect inspection,also known as defect detection,is significant in mobile screen quality control.There are some challenging issues brought by the characteristics of screen defects,including the following:(1)the problem of interclass similarity and intraclass variation,(2)the difficulty in distinguishing low contrast,tiny-sized,or incomplete defects,and(3)the modeling of category dependencies for multi-label images.To solve these problems,a graph reasoning module,stacked on a classification module,is proposed to expand the feature dimension and improve low-quality image features by exploiting category-wise dependency,image-wise relations,and interactions between them.To further improve the classification performance,the classifier of the classification module is redesigned as a cosine similarity function.With the help of contrastive learning,the classification module can better initialize the category-wise graph of the reasoning module.Experiments on the mobile screen defect dataset show that our two-stage network achieves the following best performances:97.7%accuracy and 97.3%F-measure.This proves that the proposed approach is effective in industrial applications.