IFY-Net: Detecting Surface Defects in Real-Time on Intricate Steel Samples

Shuming Zhang, Xiaohan He, Yinchi Li, Meiling Wang*, Zhenguo Liu*

*Corresponding author for this work

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

Abstract

Metal materials are widely used in the manufacturing industry. So surface defect detection of metal products is crucial. However, the problem of intricate defect samples exists in complex industrial environments. In this study, we propose Imbalance-Focused YOLO network (IFY-Net) algorithm, offering a novel solution to this issue. First, the network's receptive field is enhanced to improve the detection accuracy of scale-imbalanced samples. Furthermore, the receptive field enhancement algorithm is improved to ensure that the network focuses on scale-abnormal samples while minimizing the loss of local features. Lastly, a self-attention mechanism is introduced to improve the network's ability to identify background-similar samples. The proposed algorithm demonstrates a 4.5% improvement in mean Average Precision (mAP) on GC10-DET dataset, while achieving a detection speed of 159.96 FPS. The experimental results validate the algorithm's effectiveness and real-time capability.

Original languageEnglish
Title of host publicationProceedings of the 37th Chinese Control and Decision Conference, CCDC 2025
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages172-177
Number of pages6
ISBN (Electronic)9798331510565
DOIs
Publication statusPublished - 2025
Externally publishedYes
Event37th Chinese Control and Decision Conference, CCDC 2025 - Xiamen, China
Duration: 16 May 202519 May 2025

Publication series

NameProceedings of the 37th Chinese Control and Decision Conference, CCDC 2025

Conference

Conference37th Chinese Control and Decision Conference, CCDC 2025
Country/TerritoryChina
CityXiamen
Period16/05/2519/05/25

Keywords

  • defect detection
  • intricate defect samples
  • receptive field
  • self-attention mechanism
  • YOLO11

Fingerprint

Dive into the research topics of 'IFY-Net: Detecting Surface Defects in Real-Time on Intricate Steel Samples'. Together they form a unique fingerprint.

Cite this