Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace

Research Article

Improved Loss Function for Defect Detection of Mobile Phone Screen

Download775 downloads
  • @INPROCEEDINGS{10.4108/eai.27-8-2020.2294669,
        author={wang  zixuan and Li  Chengyuan and Zhang  Yifan and Zhang  Fan and Chang  Shuo and Li  Yilei},
        title={Improved Loss Function for Defect Detection of Mobile Phone Screen},
        proceedings={Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2020},
        month={11},
        keywords={defect detection deep learning cross-entropy loss contrastive loss siamese network},
        doi={10.4108/eai.27-8-2020.2294669}
    }
    
  • wang zixuan
    Li Chengyuan
    Zhang Yifan
    Zhang Fan
    Chang Shuo
    Li Yilei
    Year: 2020
    Improved Loss Function for Defect Detection of Mobile Phone Screen
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.27-8-2020.2294669
wang zixuan1,*, Li Chengyuan1, Zhang Yifan1, Zhang Fan2, Chang Shuo1, Li Yilei3
  • 1: Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunication, Beijing, P.R.China, 100876.
  • 2: School of Information and Communication Engineering, Beijing Information Science and Technology University
  • 3: Beijing University of Posts and Telecommunication, Beijing, P.R.China, 100876
*Contact email: 2924091689@qq.com

Abstract

Due to the excellent feature learning and representation capabilities of deep learning, the method based on deep learning for mobile phone screen defect detection is gradually being applied to industrial detection. Nowadays, the cross-entropy loss commonly used in deep learning only focuses on the differences between different classes with less intra-class differences. It leads to poor discriminantive ability of the model when the similarity between training samples is high. The contrastive loss reduces the intra-class variations and can distinguish between similar objects from different classes. Given the above analysis, we propose a Siamese network for mobile phone screen defect detection (SMSDD) using combined contrastive loss and cross-entropy loss, thereby enhancing the discriminative ability of model. Numerical results show that SMSDD achieves comparable performance.