Related Experiment Video
Updated: Jul 12, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
565
A Systematic Review on Deep Learning with CNNs Applied to Surface Defect Detection
Esteban Cumbajin1, Nuno Rodrigues1, Paulo Costa1
1Computer Science and Communications Research Centre, School of Technology and Management, Polytechnic of Leiria, 2411-901 Leiria, Portugal.
Journal of Imaging
|October 27, 2023
Summary
This review classifies surface defect detection using machine learning, focusing on convolutional neural networks (CNNs) for various industrial surfaces. Metallic surfaces are most studied, with classification being the primary task, and transfer learning widely used.
Area of Science:
- Materials Science
- Computer Science
- Industrial Engineering
Background:
- Surface defect detection is crucial in industry, yet fragmented information hinders progress.
- Machine learning, particularly convolutional neural networks (CNNs), offers advanced solutions.
- A structured overview is needed to guide research and application.
Purpose of the Study:
- To systematically review and classify machine learning-based surface defect detection methods.
- To focus on CNNs and categorize them by surface type (metal, building, ceramic, wood, special).
- To propose a novel machine learning taxonomy based on review findings.
Main Methods:
- Systematic literature review following PRISMA guidelines.
- Analysis of 59 primary studies focusing on defect types, surface categories, and CNN architectures.
- Extraction and summarization of key characteristics: surface type, problem, network, techniques, datasets, and challenges.
Main Results:
- Metallic surfaces are most prevalent (62.71%), with classification as the dominant problem (49.15%).
- Transfer learning (83.05%) and data augmentation (59.32%) are widely adopted techniques.
- Insights provided on camera choices, illumination strategies, and dataset creation for real-world applications.
Conclusions:
- The review provides a structured classification for surface defect detection using CNNs.
- Identified trends and a new taxonomy offer direction for future research.
- Facilitates efficient information retrieval for researchers and professionals in defect detection.
