Premium
P‐4.10: Simulation Algorithm of Industrial Defects based on Generative Adversarial Network
Author(s) -
Wang Anni,
Zhang Shengsen,
Chen Chunxu,
Zheng Zengqiang
Publication year - 2021
Publication title -
sid symposium digest of technical papers
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.351
H-Index - 44
eISSN - 2168-0159
pISSN - 0097-966X
DOI - 10.1002/sdtp.14538
Subject(s) - computer science , dependency (uml) , generative grammar , sample (material) , algorithm , field (mathematics) , process (computing) , set (abstract data type) , artificial intelligence , generative adversarial network , deep learning , artificial neural network , data mining , machine learning , pattern recognition (psychology) , mathematics , chemistry , chromatography , pure mathematics , programming language , operating system
Due to the application of deep learning method, defect detection in the industrial field is no longer limited to the traditional machine vision algorithm. However, the dependency of deep learning on samples and the lack of defect samples during industrial detection process are urgent problems that need to be solved. Based on Generative Adversarial Network (GAN) network, this paper proposes a defect sample simulation generation algorithm for industrial defect. Based on the characteristics of industrial defects, this paper first extracts the samples, intercepts the specific defects according to the location, and then uses the StyleGAN network training to generate the model. Then, based on the corresponding model of location information, new samples are generated. Finally, according to the generated samples and location information, we provide a typical image fusion method to generate the final samples. Through the simulation experiment of panel defects, the algorithm proposed in this paper can achieve the generation and fusion of defects, and form a new sample data set for use.