Online inspection system based on machine learning techniques: real case study of fabric textures classification for the automotive industry

• Type: Article in International Scientific Journal
• Date: 2019

Authors

Pedro Malaca (Author) Other; Luís F. Rocha (Author) FEUP; D. Gomes (Author) Other; João Silva (Author) Other; Germano Veiga (Author) FEUP

Journal

Title: Journal of Intelligent Manufacturing

Vol. 30 No. 1

Pages: 351-361

ISSN: 0956-5515

Publisher: Springer Nature

Indexing

ISI Web of Knowledge - 5 Citations

Scopus - 0 Citations

INSPEC

Other information

Authenticus ID: P-00K-QKG

DOI: 10.1007/s10845-016-1254-6

Abstract (EN): This paper focus on the classification, in real-time and under uncontrolled lighting, of fabric textures for the automotive industry. Many industrial processes have spatial constraints that limit the effective control of illumination of their vision based systems, hindering their effectiveness. The ability to overcome these problems using robust classification methods with suitable pre-processing techniques and choice of characteristics will increase the efficiency of this type of solutions with obvious production gains and thus economical. For this purpose, this paper studied and analyzed various pre-processing techniques, and selected the most appropriate fabric characteristics for the considered industrial case scenario. The methodology followed was based on the comparison of two different machine learning classifiers, ANN and SVM, using a large set of samples with a large variability of lightning conditions to faithfully simulate the industrial environment. The obtained solution shows the sensibility of ANN over SVM considering the number of features and the size of the training set, showing the better effectiveness and robustness of the last. The characteristics vector uses histogram equalization, Laws filter and Sobel filter, and multi-scale analysis. By using a correlation based method was possible to reduce the number of features used, achieving a better balanced between processing time and classification ratio.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 11