The DOI matches its Crossref record; 135 of the 135 references it cites were found in a registry or an index; 10 of the 23 sources its claims cite were checked for support.
Steel Surface Defect Recognition: A Survey
Open accessAbstract
Steel surface defect recognition is an important part of industrial product surface defect detection, which has attracted more and more attention in recent years. In the development of steel surface defect recognition technology, there has been a development process from manual detection to automatic detection based on the traditional machine learning algorithm, and subsequently to automatic detection based on the deep learning algorithm. In this paper, we discuss the key hardware of steel surface defect detection systems and offer suggestions for related options; second, we present a literature review of the algorithms related to steel surface defect recognition, which includes traditional machine learning algorithms based on texture features and shape features as well as supervised, unsupervised, and weakly supervised deep learning algorithms (Incomplete supervision, inexact supervision, imprecise supervision). In addition, some common datasets and algorithm performance evaluation metrics in the field of steel surface defect recognition are summarized. Finally, we discuss the challenges of the current steel surface defect recognition algorithms and the corresponding solutions, and our future work focus is explained.
Overview
CoChat Summary
TL;DR
This survey reviews steel surface defect recognition, covering hardware choices, traditional machine learning and deep learning algorithms, datasets, and evaluation metrics. It argues deep learning is mainstream, summarizes four key challenges—insufficient data, imbalanced data, real-time detection, and small-object detection—and proposes future work on a public evaluation platform and weakly supervised learning.
Key findings
- Hardware: digital CMOS cameras are recommended over analog/CCD cameras for steel mill lines; LED light sources are most widely used, with up to 100,000 h life and response in 10 microseconds or less; diffusers are suggested to reduce specular glare.
- Algorithm taxonomy: traditional methods are split into texture (statistical, filter, structure, model), shape, and color/grayscale; deep learning is split into supervised, unsupervised, and weakly supervised, with object detectors divided into single-stage (SSD, YOLO, RetinaNet, CenterNet) and two-stage (RCNN, Fast R-CNN, Faster R-CNN). Supervised methods generally have high precision but require heavy annotation; unsupervised methods have lower precision and unstable training; weakly supervised methods balance the two.
- Reported performance examples include SqueezeNet-based classification with ACC = 0.9750, FPS = 100, model size = 3.1 MB; CASI-Net with ACC = 0.9583, Params = 2.22 M; a YOLO-based steel strip method with ACC = 0.9755, FPS = 83, mAP = 0.9755, RECALL = 0.9586; and an improved YOLOV3 hot-rolled strip method with PRE = 0.9837, RECALL = 0.9548, F1 = 0.9690, mAP = 0.8696, FPS = 80.96.
- Datasets and metrics: the paper summarizes public datasets such as NEU (1800 grayscale images, six defect types, 300 samples each), X-SSD (1360 defect images), Severstal (12,568 grayscale images of size 1600 × 256, 4 categories), and GC10-DET (2257 images, 10 defect types, 4096 × 1000), and aggregates more than 20 performance evaluation metrics. It identifies challenges including insufficient samples, imbalanced samples, real-time detection (flat steel line about 20 m/s; long wire up to 100 m/s), and small objects (<32 × 32 pixels or area ratio <0.1).
Methods
- Design: narrative literature survey with no primary experiments or new dataset; it reviews key hardware, classifies algorithms, summarizes datasets and evaluation metrics, and discusses challenges and solutions.
- Data/sample: draws on published literature and public datasets, including NEU, X-SSD, Severstal, GC10-DET, RSDDs, DAGM, KolektorSSD2, and Kylberg Texture Dataset.
- Analysis: qualitative synthesis and tabular summaries (Tables 1–6) of methods, advantages/disadvantages, performance metrics, and dataset characteristics; no meta-analysis or statistical pooling.
Limitations
- Stated field limitations: lack of large and uniform datasets; many methods use different performance evaluation metrics; data sample insufficiency and imbalance; accuracy-speed contradiction for real-time detection; small-object detection difficulty; environmental interference; and no strict defect classification standard.
- Evident from the text: as a survey, it provides no experimental validation of its recommendations, and its performance comparisons come from heterogeneous papers with different hardware, datasets, and metrics, so the reported numbers are not directly comparable. The provided text also contains duplicated/OCR-corrupted passages and incomplete tables, which may affect completeness.
- The paper does not state limitations of the survey itself; the limitations above are either stated as field challenges or evident from the text.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
8 Valid · 5 Not valid · 9 Not checked · 32 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In addition, in the actual production environment, the interference of the environment is serious such as uneven lighting, light reflection, noise, motion blur, the existence of many false defects, and a low contrast of background clutter, resulting in the recognition effect being often less than ideal [8].”
§ Steel Surface Defect Recognition: A Survey > 1. Introduction
What the sources say
- Partly supportedcites [8] Li, Z., Tian, X., Liu, X., Liu, Y., and Shi, X. (2022). A Two-Stage Industrial Defect Detection Framework Based on Improved-YOLOv5 and Optimized-Inception-ResnetV2 Models. Appl. Sci., 12. sourceExperimental — accuracy not yet measured
Section 2 (Related Work; page unavailable) says background information can interfere with defect features and reduce recognition accuracy. The comparison of detection schemes also describes difficulty identifying fuzzy, dark, or illegible defects. The paper does not attribute poor recognition to the cited sentence’s specific conditions—uneven lighting, reflections, noise, motion blur, or false defects—or establish that recognition in actual production is often less than ideal. (LLM, conf=0.82)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The contradiction between accuracy and speed is also one of the main problems faced by the defect recognition algorithm [10]”
§ Steel Surface Defect Recognition: A Survey > 1. Introduction
What the sources say
- Partly supportedcites [10] Sun, G., Huang, D., Cheng, L., Jia, J., Xiong, C., and Zhang, Y. (2022). Efficient and Lightweight Framework for Real-Time Ore Image Segmentation Based on Deep Learning. Minerals, 12. sourceExperimental — accuracy not yet measured
The Introduction describes a speed–precision trade-off in instance segmentation, and Section 4.3.2 says improving inference speed can cost accuracy. The paper studies ore image segmentation, however, not defect recognition algorithms specifically; that application is not supported by the cited source. (LLM, conf=0.86)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Partly supported
In the paper
“most classification tasks can only find the defects with the highest confidence level in the defect category, resulting in imprecise classification results [3].”
§ Steel Surface Defect Recognition: A Survey > 1. Introduction
What the sources say
- Not checked — no full text of the sourcecites [3] He, 2019, An end-to-end steel surface defect detection approach via fusing multiple hierarchical features, IEEE Trans. Instrum. Meas. source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“so manual defect recognition can no longer meet the increasingly developed industrial practical production environment.”
§ Steel Surface Defect Recognition: A Survey > 1. Introduction
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For both methods, based on texture features and shape features, they can also be used in combination.”
§ Steel Surface Defect Recognition: A Survey > 3.1.2. Shape Feature-Based Methods
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
5 more claims
Create a free account to see all 54 claims and every source verdict.
- It may state two separate claims as one. 30
- Probably not a substantial claim. 30
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“| | 2014 | [60] | Plate steel | Defect Detection | Gabor | TPR = 0.9446, FNR = 0.29 |”
§ Steel Surface Defect Recognition: A Survey > 6. Summary and Future Work
Reported 0.29 · recomputed 0.0554
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Steel Surface Defect Recognition: A Survey > Filter Based Methods
“Gabor filter [55] Suitable for high-dimensional feature spaces with low computational burden.”
§ Steel Surface Defect Recognition: A Survey > 3.1.1. Texture Feature-Based Methods
“In 2015, Xu et al. [55] designed a scheme that introduced Shearlet transform to provide effective multi-scale directional representation”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Steel Surface Defect Recognition: A Survey > 3.3.1. Single-Stage Methods
“The most popular single-stage algorithms include SSD [24], YOLO [27], RetinaNet [28], CenterNet [33]”
§ Steel Surface Defect Recognition: A Survey > References
“24. Iandola, F.N.; Han, S.; Moskewicz, M.W.; Ashraf, K.; Dally, W.J.; Keutzer, K. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and<0.5 MB model size.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Steel Surface Defect Recognition: A Survey > 3.1.1. Texture Feature-Based Methods
“this work was able to recognize images at 63 FPS with a high detection accuracy at the same time.”
§ Steel Surface Defect Recognition: A Survey > 6. Summary and Future Work
“| | 2019 | [53] | Plate steel | Defect Classification | Local binary mode, gray histogram | ACC = 0.9440, FPS = 15.87 |”
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812
You Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803
Densely Connected Convolutional Networks
Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.243 · Cited by 48,566
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
Histograms of Oriented Gradients for Human Detection
Dalal, N., & Triggs, B. (2005) · 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) · doi:10.1109/cvpr.2005.177 · Cited by 33,584
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.81 · Cited by 32,684
Research works printed but not linked: 1
- Iandola, F.N., Han, S., Moskewicz, M.W., Ashraf, K., Dally, W.J., and Keutzer, K. (2016). SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and<0.5 MB model size. arXiv.
Cited by
Khanam, R., Hussain, M., Hill, R., & Allen, P. (2024) · IEEE Access · doi:10.1109/access.2024.3425166 · Cited by 237
“These include steel surfaces [44], rail tracks [45], photovoltaic systems [46], [47], [48], fabrics [49], [50], [51], and displays [52].”
BackgroundBai, D., Li, G., Jiang, D., Yun, J., Tao, B., Jiang, G., Sun, Y., & Ju, Z. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.107697 · Cited by 116
Zhang, H., Li, S., Miao, Q., Fang, R., Song, X., Hu, Q., Hu, J., & Chan, S. (2024) · Scientific Reports · doi:10.1038/s41598-024-57990-3 · Cited by 104
Song, K., Feng, H., Cao, T., Cui, W., & Yan, Y. (2024) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2024.3383513 · Cited by 94
“Regrettably, SSTs are prone to inner surface defects, suchas cracks, pits, scratches, andoxidescales duetoproduction technology and equipment [2].”
BackgroundSelf-citationCross Position Aggregation Network for Few-Shot Strip Steel Surface Defect Segmentation
Feng, H., Song, K., Cui, W., Zhang, Y., & Yan, Y. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3246519 · Cited by 112
“STRIP steel is an essential raw material in industrial production and manufacturing [1], [2], [3], [4], and the surface quality of strip steel will directly influence its production-grade and work performance.”
BackgroundSelf-citation
Metadata
- Authors
- Wen, X., Shan, J., He, Y., & Song, K.
- Affiliations
- Shenyang University of Technology (CN)
- Northeastern University (CN)
- Corresponding author
- Song, Kechen
- Venue
- Coatings
- Publisher
- MDPI AG
- Published
- 2022-12-22
- Volume
- 13
- Issue
- 1
- Pages
- 17
- ISSN
- 2079-6412 (ISSN-L)
- OpenAlex
- W4313327864
- Semantic Scholar
- bd464155bd36398650bff7421e6986cb707ad2f7
- Cited by
- 210
- References
- 135
- Citations by year
- 2026: 61 · 2025: 64 · 2024: 55 · 2023: 30
- FWCI
- 14.36
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Coatings (publishedVersion, cc-by)
- Funders
- Liaoning Provincial Department of Education Scientific Research Project (LQGD2020023)
- Topics
- Physical Sciences › Engineering › Industrial and Manufacturing Engineering › Industrial Vision Systems and Defect Detection
- Physical Sciences › Engineering › Mechanical Engineering › Non-Destructive Testing Techniques
- Physical Sciences › Engineering › Civil and Structural Engineering › Infrastructure Maintenance and Monitoring
- Keywords
- surface defect detection, deep learning, texture features, shape features, weakly supervised learning, defect dataset, evaluation metrics, industrial inspection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Steel Surface Defect Recognition: A Survey. https://app.cochat.ai/papers/10.3390/coatings13010017
- The paper
- Wen, X., Shan, J., He, Y., & Song, K. (2022). Steel Surface Defect Recognition: A Survey. Coatings. https://doi.org/10.3390/coatings13010017
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.