Deep Learning Based Defect Inspection Using the Intersection Over Minimum Between Search and Abnormal Regions

Eunjeong Choi, Jeongtae Kim

Research output: Contribution to journalArticlepeer-review

12 Scopus citations


We present a deep learning based defect inspection system that detects bounding boxes for any identified defect regions. In contrast to existing deep learning based object detection methods, the proposed method detects defects based on the intersection over minimum between a proposal region and defect regions rather than the well-known intersection over union, since intersection over minimum is more effective to detect variously sized defects. The proposed method also provides significant improvements over existing methods such as efficient training by minimizing cross entropy loss function, and efficient defect detection using multiple proposal boxes for the defect and entire image. We verified that the proposed method provides improved performance compared with existing methods using simulation and experimental studies.

Original languageEnglish
Pages (from-to)747-758
Number of pages12
JournalInternational Journal of Precision Engineering and Manufacturing
Issue number4
StatePublished - 1 Apr 2020

Bibliographical note

Funding Information:
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning (2017R1A2B4004231) and by the MOTIE [Ministry of Trade, Industry and Energy (10079560)] and Development of materials and core-technology for future display support program. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Publisher Copyright:
© 2020, Korean Society for Precision Engineering.


  • Deep learning
  • Defect inspection
  • Machine vision
  • Object detection


Dive into the research topics of 'Deep Learning Based Defect Inspection Using the Intersection Over Minimum Between Search and Abnormal Regions'. Together they form a unique fingerprint.

Cite this