Recent advances in surface defect inspection of industrial products using deep learning techniques

Xiaoqing Zheng,Song Zheng,Yaguang Kong,Jie Chen
DOI: https://doi.org/10.1007/s00170-021-06592-8
IF: 3.563
2021-01-25
The International Journal of Advanced Manufacturing Technology
Abstract:Manual surface inspection methods performed by quality inspectors do not satisfy the continuously increasing quality standards of industrial manufacturing processes. Machine vision provides a solution by using an automated visual inspection (AVI) system to perform quality inspection and remove defective products. Numerous studies and works have been conducted on surface inspection algorithms. With the advent of deep learning, a number of new algorithms have been developed for better inspection. In this paper, the state-of-the-art in surface defect inspection using deep learning is presented. In particular, we focus on the inspection of industrial products in semiconductor, steel, and fabric manufacturing processes. This work makes three contributions. First, we present the prior literature reviews on vision-based surface defect inspection and analyze the recent AVI-related hardware and software. Second, we review traditional surface defect inspection algorithms including statistical methods, spectral methods, model-based methods, and learning-based methods. Third, we investigate recent advances in deep learning-based inspection algorithms and present their applications in the steel, fabric, and semiconductor industries. Furthermore, we provide information on publicly available datasets containing surface image samples to facilitate the research on deep learning-based surface inspection.
What problem does this paper attempt to address?