Skip to main navigation Skip to search Skip to main content

TGNet: Texture-Guided Network for Strip Steel Surface Defect Detection

  • Sheng Gao*
  • , Honghao Wang
  • , Lianlei Lin
  • , Jiawei Wang
  • *Corresponding author for this work
  • School of Electronics and Information Engineering, Harbin Institute of Technology
  • Harbin University of Science and Technology
  • Ltd.

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Strip steel surface defect segmentation is a challenging task due to the high visual similarity between defect regions and the surrounding normal surface. Motivated by the sensitivity of texture features to subtle visual differences, as well as the success of camouflaged object detection (COD) in distinguishing visually similar patterns, this study explores the effectiveness of optimizing a deep model with texture features and proposes a texture-guided COD network (TGNet) for strip steel surface defect segmentation. In which a target texture learning module is used to guide the model to learn the texture features of the camouflaged object. A texture guiding module integrates target texture features and features extracted by the feature extractor. An adjacent fusion module integrates features from adjacent layers. The experimental results obtained on the four datasets demonstrate that the texture-guided method outperformed the other state-of-the-art methods for both strip steel surface defect segmentation and naturally camouflaged object detection.

Original languageEnglish
Title of host publication2025 Global Reliability and Prognostics and Health Management Conference, PHM-Xian 2025
EditorsHuimin Wang, Steven Li
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798331526757
DOIs
StatePublished - 2025
Externally publishedYes
Event16th IEEE Reliability and Prognostics and Health Management Conference, PHM-Xian 2025 - Xian, China
Duration: 10 Oct 202512 Oct 2025

Publication series

Name2025 Global Reliability and Prognostics and Health Management Conference, PHM-Xian 2025

Conference

Conference16th IEEE Reliability and Prognostics and Health Management Conference, PHM-Xian 2025
Country/TerritoryChina
CityXian
Period10/10/2512/10/25

Keywords

  • camouflaged object detection
  • deep learning
  • defect segmentation
  • strip steel
  • texture feature

Fingerprint

Dive into the research topics of 'TGNet: Texture-Guided Network for Strip Steel Surface Defect Detection'. Together they form a unique fingerprint.

Cite this