Effective and Lightweight Surface Defect Detection via Enhanced Deep Learning Models
This study addresses limitations in industrial defect detection by enhancing YOLOv8 with EfficientNet-B0. Using MBConv, multiscale pathways, and C2f modules, the model improved precision from 0.935 to 0.983, the recall from 0.917 to 0.981, [email protected] from 0.964 to 0.993, and [email protected]:0.95 from 0.787 to 0.9 on the NEU-DET dataset, enabling efficient, real-time defect detection for industrial applications.
Authors
- Ji Zhang (ORCID: https://orcid.org/0000-0002-1244-2880)
- Hong-Zhou Li (ORCID: https://orcid.org/0000-0003-3118-1599)
- Kenny Pierrot Kayanzari
Institutions
- University of Southern Queensland (AU)
- Guilin University of Electronic Technology (CN)
- Nanjing University of Aeronautics and Astronautics (CN)
Publication Details
- Published
- 2024-12-15
- DOI
- https://doi.org/10.1109/bigdata62323.2024.10825832
- Primary Topic
- Industrial Vision Systems and Defect Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00