Deep learning for industrial image: challenges, methods for enriching the sample space and restricting the hypothesis space, and possible issue

ABSTRACT

Deep learning (DL) is an important enabling technology for intelligent manufacturing. The DL-based industrial image pattern recognition (DLBIIPR) plays a vital role in the improvement of product quality and production efficiency. Although DL technology has been widely used in the field of natural image, industrial image often has some mixed characteristics, such as small sample, imbalance, small target, strong interference, fine-grained, temporality and semantical, which reduce the feasibility and generalization of DLBIIPR. To solve this problem, this paper provides an overview of approaches commonly used in industry by enriching the sample space and limiting the hypothesis space. In order to improve the confidence of front-line workers in using DL models, the explainable deep learning (XDL) methods are reviewed, and a case study is used to verify the effectiveness of XDL.

KEYWORDS:

• Intelligent manufacturing
• deep learning
• industrial Image
• pattern recognition
• over-fitting
• computational learning theory
• sample space
• hypothesis space
• explainable deep learning

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China [51905091]; The Fundamental Research Funds for the Central Universities and Graduate Student Innovation Fund of Donghua University [CUSF-DH-D-2020053].