Scribd
Upload
24 views2 pages

Steel Surface Defect Classification W Deep Learning

Uploaded by

efemurathansahin13
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
24 views2 pages

Steel Surface Defect Classification W Deep Learning

Uploaded by

efemurathansahin13
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Steel Surface Defect Classification Via Deep Learning

Mehmet Talha Bozan, Mustafa Mert Tunali, Ahmet Yildiz


Computer Engineering Department, MEF University, Istanbul, Turkey

1. INTRODUCTION c. Model Architectures:


• Artificial Intelligence (AI) and Deep Learning (DL) continue to develop every day This study uses three state-of-the-art Convolution Neural
and take more place in our lives. The use of AI applications in quality control has Networks (CNN) architectures:
increased rapidly and can help to improve industry [1]. • InceptionV3 [5]
• Xception [6] (Figure 6 and 7)
• The use of AI applications in quality control reduces production costs [2]. • VGG [7]

• Quality control has been tried to be achieved by using heuristic image processing Parameters of model:
algorithms [3]. • Pre-trained weights initialized with ImageNet [8]
• Transfer learning
• The purpose of this article is to improve quality control, that is, to deliver excellent • All layers was frozen except Batch-Normalization layers
and super quality products to consumers, using deep learning methods with NEU- • Adam optimizer were used with learning rate of 0.001
DET dataset. • Categorical cross-entropy as a loss function
Figure 6. Depth-wise and Pointwise Convolution
2. MATERIALS & METHODS representation of Xception [9]
Figure 8. ROC Curve of the model

a. Data Collection and Pre-processing


§ 6 classes such as crazing, inclusion, patches, pitted surface, rolled-in scale and
scratches (Figure 1).
§ 1800 images (80% training, 20% validation)
§ 200 x 200 image resolutions
§ Rescaling (0-255 to 0-1)

Crazing Inclusion Patches Figure 9. The prediction results on validation dataset

Figure 7. Xception Architecture [9] 4. DISCUSSION & CONCLUSION


d. Metrics
• Receiver Operationg Characteristics’s (ROC) and Area Under Curve (AUC) (Equation 1) Object detection can be difficult. Thus, working on our problem would be fit for the
• Categorical Accuracy (ACC) (Equation 2) industry. For example, using models with minimal computing power can help to
• ROC curve was calculated in a one-vs-rest way reduce server pricing and complexity.

The aim of this study was to detect deformations on steel surfaces with deep learning

Pitted Surface Rolled-in Scale Scratches


𝐴𝑈𝐶 = ∫ 𝑇𝑃𝑅 𝑑 𝐹𝑃𝑅 (1) techniques and to increase quality control. The main finding in this study is the
detection of deformations on steel surfaces with a 95.65% success rate using the
Xception model architecture. Future studies can be aimed to detect on the real-time
Figure 1. Example collected images from NEU-DET [4] !"#!$
A𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (2) production line.
(!"#&"#!$#&$)
b. Data Augmentation References
• Random flip with horizontal and vertical axis (Figure 3)
• Rotation with 20 degrees clockwise and counterclockwise (Figure 4 and 5)
3. RESULTS
• Model training results are in Table 1, ROC is in Figure 8, and Confusion matrix is in Figure 9.

Table 1. Model evaluation table


Training Validation
Model AUC
Accuracy Accuracy
VGG19, w/Adam @ 0.001 LR 0.926 0.32 0.53

InceptionV3, w/Adam @ 0.001 LR 0.9643 0.9454 0.9767


Figure 2. Figure 3. Figure 4. 20 Figure 5. 20 degrees
Original Image Horizontal degrees clockwise counterclockwise Xception, w/Adam @ 0.001 LR 0.9755 0.9565 1.0
[4] flipped [4] rotation [4] rotation [4]
REFERENCES

[1] P. Santhanam, “Quality Management of Machine Learning Systems,” Jun. 2020. [Online]. Available: https://arxiv.org/pdf/2006.09529.

[2] A. GORCHET, “Deep Learning Has Reinvented Quality Control in Manufacturing—but It Hasn’t Gone Far Enough,” IEEE Spectrum, Nov. 20, 2020. https://spectrum.ieee.org/deep-
learning-has-reinvented-quality-control-in-manufacturingbut-it-hasnt-gone-far-enough (accessed Nov. 07, 2021).
[3] X. Lv, F. Duan, J. Jiang, X. Fu, and L. Gan, “Deep Metallic Surface Defect Detection: The New Benchmark and Detection Network,” Sensors, vol. 20, no. 6, p. 1562, Mar. 2020, doi:
10.3390/s20061562.
[4] K. Dixit, NEU Surface Defect Database, ver. 1, Kaggle 2020. [Dataset] Available: https://www.kaggle.com/kaustubhdikshit/neu-surface-defect-database.
[5] Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z. Rethinking the Inception Architecture for Computer Vision. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR) [Internet]. Las Vegas, NV, USA: IEEE; 2016 [cited 2021 Jul 26]. p. 2818–26. Available from: http://ieeexplore.ieee.org/document/7780677/

[6] Chollet F. Xception: Deep Learning with Depthwise Separable Convolutions. In: 2017 IEEE Conferenceon Computer Vision and Pattern Recognition (CVPR). 2017. p. 1800–7.

[7] Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-Scale Image Recognition. ArXiv14091556 Cs[Internet]. 2015 Apr 10 [cited 2021 Sep 4]; Available from:
http://arxiv.org/abs/1409.1556

[8] J. Deng, W. Dong, R. Socher, L.-J. Li, Kai Li, and Li Fei-Fei, “ImageNet: A large-scale hierarchical image database,” 2009 IEEE Conference on Computer Vision and Pattern Recognition,
Jun. 2009, doi: 10.1109/cvpr.2009.5206848.

[9] nikan, “FCN or CNN, AlexNet, VGG, ResNet, Inception(GoogleNet), Xception and CIFAR10 classifier,” IUST Projects, Feb. 14, 2020. https://iust-projects.ir/post/dip10/ (accessed Jan. 03,
2022).

You might also like