Skip to main content

Table 3 Comparison of defect detection systems

From: Review of vision-based steel surface inspection systems

Paper

Method

Type of defects

Sample size

Features

Detection accuracy (%)

Resolution (across × along)

Speed of steel object (m/s)

Real-time operation

Remark

Detection

Classification

[67] - slab

Gabor filter, two-level thresholding, edge pair detection

SVM

Scratch

7,110 cases

7 histogram, gradient

94.08

   

Classification w.r.t. pseudo defect

[29] - slab

Gabor filter, adaptive double-thresholding

Feature-based logic

Pinhole

1,764 images

4 morphological features

87.1

0.57 × 0.5 mm

  

Classification w.r.t. pseudo defect

[38] - billet

DWT, morphological

Feature difference

Corner crack

1,568 regions

4 Morphological

97.6

   

Classification w.r.t. pseudo defect

[64] - billet

Wavelet reconstruction, double threshold

SVM

Corner crack

220 images

12 Histogram, morphological

97.8

0.25 mm along web

2

Suitable

Classification w.r.t. pseudo defect

[65] - plate

Gabor filter, adaptive thresholding

SVM

Seam crack

10,459 images

12 geometric, gray

84.83

0.5 mm

  

Classification w.r.t. pseudo defect

[41] - plate

UDWT (undecimated WT), morphology

 

Crack, scratch

563 images

 

90.23

    

[24] - hot strip

Background difference, region growing

 

Scar, scratches, pits, cracks.

8,037 defects

 

>90

0.5 × 0.5 mm

10

Suitable

Bright and dark mode cameras used

[42] - cold strip

Morphological, curvelet

Linear discriminant analysis

 

800 images

13 curvelet +7 morphological

98.5

1× 1 mm

  

Pixel-level classification as defect

[56] - cold strip

Background difference, gray-level distribution

 

Wrinkles, inclusion, weld, holes and serrated edges

150 images

 

99.3

    

[35] - cold strip

Multivariate discriminant function

 

Wrinkles

40 images

 

91

  

Suitable

 

[3] - cold strip

Hough transform-hole, scratch. Double thresholding-coil break. Renyi entropy- rust

 

Hole, scratch, Coil break and rust

93 + 157 synthesized images

 

78 to 90.4

    

[61] - rod/bar

Edge preserving filter, double threshold

 

Crack, spot, dark line

175 defects (73 images)

 

95.42

 

18.5

Suitable

 

[58] - rod/bar

Local annular contrast

 

Pits, overfill, scratch

408 images

 

93.88 to 100

 

4.6

Suitable

 

[60] - rod/bar

Sobel edge detector, snake projection, DWT

T2 control chart

Seam

400 subimages

7 to 9

97.5

 

Approximately 18

Suitable

Classification w.r.t. pseudo defect

[44] - rod/bar

Gradient filter, double thresholding,

SVM

Vertical scratch

2,444 images

42 geometric, gray level

96.9

0.3 mm

18.5

 

Five cameras

Classification w.r.t. pseudo defect

[2] - rod/bar

Gradient filter, region growing,

SVM-RBF

Seam

1,226 images

Geometric, gray level

94.4

 

100

Suitable

Classification w.r.t. pseudo defect

[5] - rod/bar

UDWT-(Haar), double threshold

SVM

Scratch

2,080 data

14 geometric, gray level

91.83

0.5 mm

18

 

Five cameras

Classification w.r.t. pseudo defect

[75] - rod/bar

UDWT (Haar), DFT

 

Periodic defects

6 coils

 

100

0.5 mm along web

18

Suitable

Classification w.r.t. pseudo defect

[59] - rod/bar

Special horizontal, vertical, diagonal edge filters

 

Seam, scratch, roll mark, overfill

663 images

12 geometric, gray level

85.82 to 89

 

15

Suitable

Â