The performance comparison of optical character recognition and object detection methods on rubber products

Abstract

In the manufacturing sector, there are dedicated codes or characters on finished and semi-finished products that are specifically defined by manufacturers. These codes are composed of characters that have special meanings in themselves and they are usually made up of different combinations of numbers and letters. This study aims to detect the engraved letter and number combinations from the surfaces of cured rubber specimens and rubber products. The data set has been prepared from products and specimens in the production line where the controlling system will be set. The study covers two different approaches which are OCR and object detection. To decide which method satisfies the requirements, detailed analyzes were made along with the detection performances and their robustness against scene changes that may occur due to the nature of the problem.  Based on the findings, the performance of OCR methods has not been found as satisfactory.

Keywords: Deep Learning; Low Contrast; Object Detection; OCR; Rubber Sample.