| ■■■■■■■■■■■■■■■■■■■■ KAWASAKI STEEL TECHNICAL REPORT No.26 ( June 1992 ) Artificial Intelligence and Wire Rods and Steel Bars ■■■■■■■■■■■■■■■■■■■■ |
| Neural Network Model for the Recognition of Characters Stenciled on Continuously Cast Slabs Kazuya Asano, Jun-ichi Tateno, Satoshi Maruyama, Kazuo Arai, Michio Ibaraki, Masaru Shibata |
|
Synopsis : The No. 3 continuous caster at Chiba Works of Kawasaki Steel employs identification cokes that are stenciled on the surface of cast slabs. The stenciled characters are occasionally degraded by blurs or spots and may become unreadable for the operators who handle the slabs. In order to detect such defects in the characters and to assure quality of marking, a character recognition system was developed using a three-layer neural network model. Characters are extracted from an image by a pattern matching method. After preprocessing, the images of each character are provided to the neural network model, which produces a criterion of similarity between the character images and standard patterns. An alarm is given to the operators when the quality of a character is judged to be unacceptable by the neural network model. The ability to correctly classify the characters is more than 99%, and the system is being successfully used to assure the quality of slab marking. |
| *Copyright folder's permission is being asked for the publication of the report. Publication of the report on the web is temporally suspended. Please ask libraries which hold KAWASAKI STEEL TECHNICAL REPORT for a copy of the full text. |