Image Identification for Surface Defects of Steel Ball Based on Support Vector Machine

Abstract:

In response to the dilemma for image identification by the existing classifier toward surface defects of steel ball, an improved support vector machine (SVM) for multiclass problems is proposed. Minimum distance method is presented to resolve the unclassifiable region of the multiclass SVMs. The 16 image features of the surface defects are selected as input vector of the SVMs. The experiment results show that more accurate identification toward surface defects of steel ball was achieved by the improved multiclass SVM and the accuracy can reach 95%.

• Periodical: Advanced Materials Research (Volumes 199-200)
• Pages: 1769-1772
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.199-200.1769
• Online since: February 2011
• Authors: Yong Wei Yu, Guo Fu Yin, Liu Qing Du
• Keywords: Image Identification, Steel Ball, Support Vector Machine (SVM), Surface Defect

References

[1] Cortes C, Vapnik V: Machine Learning Vol. 20 (1995), p.273.

[2] Smola AJ, Scholkopf B: Statistics and Computing Vol. 14 (2004), p.199.

[3] Tipping ME; Journal of Machine Learning Research Vol. 1 (2001), p.211.

[4] Hwei Jen Lin, Jih Pin Yeh: Applied Mathematics and Computation Vol. 214 (2009), p.329.

[5] Shinya Katagiri, Shigeo Abe: Pattern Recognition Letters Vol. 27 (2006), p.1495.

[6] Ming-Huwi Horng: Expert Systems with Applications Vol. 36 (2009), p.8124.

[7] Sumeet Agarwal, V. Vijaya Saradhi, Harish Karnick: Neurocomputing Vol. 71 (2008), p.1230.

[8] Chih-Wei Hsu and Chih-Jen Lin; IEEE Transaction on Neural Networks Vol. 13 (2002), p.415.

DOI: 10.1109/72.991427