Publication detail

Visual defect detection in real-world industrial applications using convolutional neural networks

JEŽEK, Š.

Original Title

Visual defect detection in real-world industrial applications using convolutional neural networks

Type

article in a collection out of WoS and Scopus

Language

English

Original Abstract

This paper presents a benchmark of current deep learning-based visual anomaly detection methods. Anomaly detection is a critical problem in various field such as fraud detection, cybersecurity and also industrial quality control. Modern advancements in the area of deep learning have shown a significant promise in the ability to automatically detect defects in the image data and this can be exploited in many industrial applications. With the recent development in the field of deep learning, many new visual anomaly detection methods have been introduced recently. These methods are usually evaluated using datasets that do not cover several industrial use cases very well. Performance of many methods may therefore differ, when used in new applications. In this paper we perform the image-level and pixel-level AUROC evaluation of recent state of the art visual anomaly detection methods using the MPDD dataset that aims to simulate complex conditions that can be encountered during data acquisition in metal parts manufacturing. These conditions include changes in rotation and position of objects to the camera, changing number of parts in the image or limited number of training samples. We show that recent state of the art mathods may have a significantly lower precision (by up to 30 % AUROC) when deployed in several industrial use cased evaluated in our benchmark.

Keywords

visual anomaly detection, industrial visual in-spection, convolutional neural networks, image processing

Authors

JEŽEK, Š.

Released

25. 4. 2023

Location

Brno

ISBN

978-80-214-6153-6

Book

Proceedings I of the 29th Student EEICT 2023 (General Papers)

Pages from

389

Pages to

393

Pages count

5

URL

https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2023_sbornik_1.pdf

BibTex

@inproceedings{BUT183929,
  author="Štěpán {Ježek}",
  title="Visual defect detection in real-world industrial applications using convolutional neural networks",
  booktitle="Proceedings I of the 29th Student EEICT 2023 (General Papers)",
  year="2023",
  pages="389--393",
  address="Brno",
  isbn="978-80-214-6153-6",
  url="https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2023_sbornik_1.pdf"
}