JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

PEYOLO a perception efficient network for multiscale surface defects detection

  • 0Xi'an Key Laboratory of Advanced Photo-Electronics Materials and Energy Conversion Device, School of Electronic Information, Xijing University, Xi'an, 710123, People's Republic of China.
Scientific Reports +

|

August 6, 2025

|

August 6, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces PEYOLO, an efficient network for detecting small steel surface defects. PEYOLO improves accuracy and speed in complex production environments for real-time quality control.

Area Of Science

  • Materials Science
  • Computer Vision
  • Artificial Intelligence

Background

  • Steel defect detection is vital for quality control in manufacturing.
  • Detecting small-scale defects in complex industrial settings presents significant challenges.
  • Existing methods struggle with multi-scale defects and efficiency.

Purpose Of The Study

  • To develop a perception-efficient network for fast and accurate detection of multi-scale steel surface defects.
  • To enhance feature fusion and global feature capture for improved defect identification.
  • To provide a solution suitable for real-time steel defect detection applications.

Main Methods

  • Introduced Defect Capture Path Aggregation Network for multi-scale feature learning.
  • Designed Perception-Efficient Head (PEHead) to reduce missed detections by mitigating aliasing.
  • Proposed Receptive Field Extension Module (RFEM) to enhance global feature capture and handle aspect ratio variations.
  • Integrated these modules into the YOLO framework, creating PEYOLO.

Main Results

  • PEYOLO achieved mAP50 improvements of 3.5% (NEU-DET), 9.1% (GC10-DET), and 3.3% (Severstal) over YOLOv8n.
  • The method maintained high inference speed, comparable to real-time requirements.
  • Demonstrated significant performance gains on public steel defect datasets.

Conclusions

  • PEYOLO effectively addresses the challenge of detecting small-scale, multi-scale steel surface defects.
  • The proposed network offers a balance of high accuracy and fast inference speeds.
  • PEYOLO is a viable solution for real-time steel defect detection in industrial quality control.

Keywords: