EU‐Net: A novel semantic segmentation architecture for surface defect detection of mobile phone screens

Abstract Manual or conventional image processing algorithms are commonly used to detect surface problems on mobile phone screens. However, inefficiency and inflexibility are disadvantages. Although the semantic segmentation method has high adaptability and accuracy, it also has a low defect detection efficiency due to its excessive parameters. In order to increase defect detection efficiency, a novel efficient encoder–decoder architecture termed MB encoder–decoder architecture based on MBConv blocks, and that it reduces the number of parameters used in semantic segmentation methods i presented. In addition, by applying the MB encoder–decoder design to the U‐Net, the efficient U‐Net (EU‐Net) is proposed. It confirms the MB encoder–decoder architecture's superiority. Then, EU‐Net to mobile phone surface defect detection in real industrial scenarios. Experimental results on a dataset show the superiority of the proposed algorithm and it can meet the real‐time requirement of industrial production.