Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 46 of the 46 references it cites were found in a registry or an index; 11 of the 13 sources its claims cite were checked for support.

X-SDD: A New Benchmark for Hot Rolled Steel Strip Surface Defects Detection

Open access
Feng, Xinglong, Gao, Xianwen, Luo, Ling · 2021 · Symmetry

Abstract

It is important to accurately classify the defects in hot rolled steel strip since the detection of defects in hot rolled steel strip is closely related to the quality of the final product. The lack of actual hot-rolled strip defect data sets currently limits further research on the classification of hot-rolled strip defects to some extent. In real production, the convolutional neural network (CNN)-based algorithm has some difficulties, for example, the algorithm is not particularly accurate in classifying some uncommon defects. Therefore, further research is needed on how to apply deep learning to the actual detection of defects on the surface of hot rolled steel strip. In this paper, we proposed a hot rolled steel strip defect dataset called Xsteel surface defect dataset (X-SDD) which contains seven typical types of hot rolled strip defects with a total of 1360 defect images. Compared with the six defect types of the commonly used NEU surface defect database (NEU-CLS), our proposed X-SDD contains more types. Then, we adopt the newly proposed RepVGG algorithm and combine it with the spatial attention (SA) mechanism to verify the effect on the X-SDD. Finally, we apply multiple algorithms to test on our proposed X-SDD to provide the corresponding benchmarks. The test results show that our algorithm achieves an accuracy of 95.10% on the testset, which exceeds other comparable algorithms by a large margin. Meanwhile, our algorithm achieves the best results in Macro-Precision, Ma…

Overview

CoChat Summary

TL;DR

The paper introduces X-SDD, a hot-rolled steel strip surface defect dataset with seven defect types and 1,360 images, and benchmarks deep learning classifiers on it. It reports that RepVGG combined with spatial attention achieves 95.10% accuracy and the best Macro-Precision, Macro-Recall, and Macro-F1 among compared models.

Key findings

  • X-SDD contains 1,360 defect images at 128 × 128 pixels in 3-channel JPG, covering seven defect types: 238 slag inclusions, 397 red iron sheet, 122 iron sheet ash, 134 surface scratches, 63 oxide scale of plate system, 203 finishing roll printing, and 203 oxide scale of temperature system. It has one more category than NEU-CLS (six types) and is imbalanced, with the largest class more than 6 times the smallest.
  • RepVGG+SA achieves 95.10% accuracy, 93.92% Macro-Recall, 95.16% Macro-Precision, and 93.25% Macro-F1 on the 408-image test set; single RepVGG_B3g4 achieves 91.67% accuracy. Across compared algorithms, accuracy ranges 87.01–95.10%, Macro-Recall 82.04–93.92%, Macro-Precision 82.04–95.16%, and Macro-F1 81.58–93.25%.
  • Per-class accuracy for RepVGG+SA: finishing roll printing 100%, oxide scale of plate system 78.95% (lowest), red iron sheet 94.12%, scratches 97.50%, inclusion 98.36%, iron sheet ash 83.78%, and oxide scale of temperature system 98.36%.
  • Model complexity: RepVGG+SA has 83.825M parameters and 17.892 G MACs, compared with EspNet-v2 at 0.627M parameters and 0.090 G MACs.

Methods

  • Dataset was collected from the hot rolled steel strip field; 70% was used as trainset (952 images) and 30% as testset (408 images). Images were resized to 224 × 224 pixels; training used 100 epochs, mini-batch 10, learning rate 0.0001, and Adam optimization on a single NVIDIA RTX2080S GPU, Intel Core i7-9700 CPU, 16GB RAM, Windows 10, and PyTorch.
  • The proposed model is RepVGG_B3g4 combined with a spatial attention mechanism (RepVGG+SA), using RepVGG structural re-parameterization and a spatial attention module. It is compared against EspNet-v2, GhostNet, ShuffleNet, SqueezeNet, Xception, VGG16, ResNet50/101/152, RepVGG_B1g2, and RepVGG_B3g4.
  • Evaluation metrics are Accuracy, Macro-Recall, Macro-Precision, and Macro-F1; macro metrics are obtained by treating the multiclass problem as multiple binary classification problems and averaging per-class Recall, Precision, and F1.

Limitations

  • Stated: RepVGG+SA performs poorly on oxide scale of plate system, with only 78.95% accuracy; the authors attribute this to the relatively small sample size for that category and its morphological similarity to slag inclusion, suggesting more samples, cascade structure, or artificial prior knowledge as possible improvements.
  • Stated: RepVGG+SA is computationally costly, with 83.825M parameters and 17.892 G MACs, much larger than EspNet-v2 (0.627M, 0.090 G); future work aims to reduce computational complexity.
  • Evident from the text: X-SDD is imbalanced (largest class more than 6 times the smallest) and relatively small (1,360 images from a single hot-rolled strip source); no cross-dataset training or evaluation on NEU-CLS is reported, only dataset comparison.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/sym13040706

Cites

46 of 46 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Coils of steel strip.
Figure 1 Coils of steel strip.CC-BY · From the PDF
Figure 2 Machine vision defect detection system.
Figure 2 Machine vision defect detection system.CC-BY · From the PDF
Figure 3 Samples of seven kinds of typical surface on X-SDD. (a-inclusion, b-red iron sheet, c-iron sheet ash, d-scratches, e-oxide scale of plate system, f-finishing roll printing and g-oxide scale of temperature system). Figure 3. Samples of seven kinds of typical surface on X-SDD. (( a )-inclusion, ( b )-red iron sheet, ( c )-iron sheet ash, ( d )-scratches, ( e )-oxide scale of plate system, ( f )-finishing roll printing and ( g )-oxide scale of temperature system).
Figure 3 Samples of seven kinds of typical surface on X-SDD. (a-inclusion, b-red iron sheet, c-iron sheet ash, d-scratches, e-oxide scale of plate system, f-finishing roll printing and g-oxide scale of temperature system). Figure 3. Samples of seven kinds of typical surface on X-SDD. (( a )-inclusion, ( b )-red iron sheet, ( c )-iron sheet ash, ( d )-scratches, ( e )-oxide scale of plate system, ( f )-finishing roll printing and ( g )-oxide scale of temperature system).CC-BY · From the PDF
Figure 4 Samples of six kinds of typical surface on NEU-CLS. (a-crazing, b-inclusion, c-patches, d-pitted surface, e-rolled in scale, f-scrathes). Figure 4. Samples of six kinds of typical surface on NEU-CLS. (( a )-crazing, ( b )-inclusion, ( c )-patches, ( d )-pitted surface, ( e )-rolled in scale, ( f )-scrathes).
Figure 4 Samples of six kinds of typical surface on NEU-CLS. (a-crazing, b-inclusion, c-patches, d-pitted surface, e-rolled in scale, f-scrathes). Figure 4. Samples of six kinds of typical surface on NEU-CLS. (( a )-crazing, ( b )-inclusion, ( c )-patches, ( d )-pitted surface, ( e )-rolled in scale, ( f )-scrathes).CC-BY · From the PDF
Figure 5 The number of defects.
Figure 5 The number of defects.CC-BY · From the PDF
Figure 6 The Sketch of RepVGG architecture. Figure 6. The Sketch of RepVGG architecture.
Figure 6 The Sketch of RepVGG architecture. Figure 6. The Sketch of RepVGG architecture.CC-BY · From the PDF
Image 7
CC-BY · From the PDF
Figure 8 The spatial attention module.
Figure 8 The spatial attention module.CC-BY · From the PDF
Image 9
CC-BY · From the PDF

Verification

Claims the paper makes

7 Valid · 5 Not valid · 2 Not checked · 54 Own results and interpretations

36 more claims

Create a free account to see all 68 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 13
  • It may state two separate claims as one. 15
  • The quote only partly states this claim. 3

Create a free account to see these claims.

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 · cited 10×

    “In addition, ResNet50 performs better on X-SDD than the deeper ResNet101 and 152, indicating that the deeper the network level is not better when the amount of data is not particularly large.”
    BackgroundInfluential
  • Attention Is All You Need

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404

  • 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 · cited 1×

    “…defect classification networks tend to use off-the-shelf deep learning network structures and their various variants, including AlexNet [23], VGGNet [24], GoogleNet [25], ResNet [26], DenseNet [27], SENet [28], Shuf-fleNet [29] and MobileNet [30], etc. Compared with traditional algorithms such as…”
    Method
  • 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

  • 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 · cited 2×

    “In general, existing strip surface defect classification networks tend to use off-the-shelf deep learning network structures and their various variants, including AlexNet [23], VGGNet [24], GoogleNet [25], ResNet [26], DenseNet [27], SENet [28], Shuf-fleNet [29] and MobileNet [30], etc. Compared with traditional algorit…”
    Background
  • 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 · cited 2×

    “In general, existing strip surface defect classification networks tend to use off-the-shelf deep learning network structures and their various variants, including AlexNet [23], VGGNet [24], GoogleNet [25], ResNet [26], DenseNet [27], SENet [28], Shuf-fleNet [29] and MobileNet [30], etc. Compared with traditional algorit…”
    Background
  • An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al. (2020) · arXiv · doi:10.48550/arxiv.2010.11929 · Cited by 47,472

  • Squeeze-and-Excitation Networks

    Hu, J., Shen, L., & Sun, G. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00745 · Cited by 35,667 · cited 2×

    “In general, existing strip surface defect classification networks tend to use off-the-shelf deep learning network structures and their various variants, including AlexNet [23], VGGNet [24], GoogleNet [25], ResNet [26], DenseNet [27], SENet [28], Shuf-fleNet [29] and MobileNet [30], etc. Compared with traditional algorit…”
    Background
  • CBAM: Convolutional Block Attention Module

    Woo, S., Park, J., Lee, J.-Y., & Kweon, I. S. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01234-2_1 · Cited by 25,853 · cited 3×

    “According to [40], max pooling and average pooling are used in channel dimension to get two different feature descriptions FS avg ∈ R1×H×W and FS max ∈ R1×H×W .”
    BackgroundMethodInfluential
  • MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., & Adam, H. (2017) · arXiv · doi:10.48550/arxiv.1704.04861 · Cited by 19,015

Show all 40 references

Research works printed but not linked: 4

  1. Aldunin, 2017, Development of method for calculation of structure parameters of hot-rolled steel strip for sheet stamping, J. Chem. Technol. Metall., 52, 737
  2. (2021, January 18). Severstal is Mastering the Production of Video Inspection Systems for Rolled Surfaces. Available online: https://metallurgprom.org/en/press-releases/4952-severstal-osvaivaet-izgotovlenie-sistem-videoinspekcii-poverhnosti-prokata.html.
  3. Zaghdoudi, R., Seridi, H., and Boudiaf, A. (2019, January 24–25). Multiple classifier combination for steel surface inspection. Proceedings of the 2nd Conference on Informatics and Applied Mathematics, IAM 2019, Guelma, Algeria.
  4. Amid, 2012, Enhanced performance for support vector machines as multi-class classifiers in steel surface defects detection, World Acad. Sci. Eng. Technol., 6, 1096

Cited by

See all 158

Metadata

Authors
Feng, X., Gao, X., & Luo, L.
Affiliations
Northeastern University (CN)
Corresponding author
Gao, Xianwen
Venue
Symmetry
Publisher
MDPI AG
Published
2021-04-17
Volume
13
Issue
4
Pages
706
ISSN
2073-8994 (ISSN-L)
OpenAlex
W3153037112
Cited by
158
References
40
Citations by year
2026: 22 · 2025: 27 · 2024: 50 · 2023: 29 · 2022: 23 · 2021: 7
FWCI
14.10
Type
journal article
Language
en
Open access
Locations
  • Symmetry (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Natural Science Foundation of China (61573087, 61573088)
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
hot-rolled strip, surface defect detection, defect classification, benchmark dataset, RepVGG, spatial attention mechanism, steel quality 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: X-SDD: A New Benchmark for Hot Rolled Steel Strip Surface Defects Detection. https://app.cochat.ai/papers/10.3390/sym13040706
The paper
Feng, X., Gao, X., & Luo, L. (2021). X-SDD: A New Benchmark for Hot Rolled Steel Strip Surface Defects Detection. Symmetry. https://doi.org/10.3390/sym13040706

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.