THE USAGE OF ARTIFICIAL INTELLIGENCE IN RECOGNITION OF EMBOSSED NUMBERS ON BILLET

1 SVOBODOVÁ Petra
Co-authors:
1 TOMEČEK Antonín 1 KRÁTKÝ Jan 1 ŠPAČKOVÁ Hana 1 KLUS Miroslav
Institution:
1 VSB - Technical University of Ostrava, Ostrava, Czech Republic, EU, petra.svobodova@vsb.cz
Conference:
30th Anniversary International Conference on Metallurgy and Materials, Brno, Czech Republic, EU, May 26 - 28, 2021
Proceedings:
Proceedings 30th Anniversary International Conference on Metallurgy and Materials
Pages:
1311-1316
ISBN:
978-80-87294-99-4
ISSN:
2694-9296
Published:
15th September 2021
Proceedings of the conference have already been published in Scopus and we are waiting for evaluation and potential indexing in Web of Science.
Metrics:
535 views / 399 downloads
Abstract

The article is focused on the recognition and localization of numbers on steel billets. Serial numbers are embossed to each billet. Our automated solution allows product identification without human interaction. There are several problems caused by embossing to the hot steel. First, the numbers are not clearly visible. There is a lot of noise around the serial number which causes shadows and reflections. Next, the surface of the billet is rough with grooves and ridges. These issues affect object detection. As a part of the 4th Industrial Revolution, artificial intelligence and neural networks are used to automate production. Object recognition identifies which numbers are presented in the image. Another problem occurs when the serial number is located anywhere on the billet surface. The aim is to detect multiple objects in the scene using a single neural network. Our proposed solution is based on an extremely fast and accurate model from a class of deep learning algorithms. To localize and identify individual numbers, the You Only Look Once (YOLO) algorithm is implemented. It predicts bounding boxes and assigns classes from category 0 – 9. The approach is fully automatic and detects embossed numbers in real time. A custom dataset and annotations for train the model is created. Due to the lack of training images, data augmentation is used to extend a dataset by increasing the amount of data.

Keywords: Artificial intelligence, recognition of embossed numbers, object detection, YOLO

© This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Scroll to Top