A Deep Learning-Based OCR System Implementation for Traceability Ensurement in a Metal Manufacturing Workshop

The traceability of all elements within a metal manufacturing industry constitutes a critical challenge in the digitalization of industrial processes. In this context, the labeling of these elements, coupled with optical character recognition (OCR), plays a significant role. Therefore, in this paper, it is accomplished this challenge by applying three pre-trained deep learning OCR models such as Pytesseract, EasyOCR and KerasOCR to a dataset of images featuring labels with corresponding alphanumeric codes. The performance of these OCR engines has been evaluated using various types of input images, to which preprocessing techniques have been applied to enhance their quality and the legibility of the text contained within. The implementation of these models produced successful results, presenting a viable solution to improve the efficiency and accessibility of information retrieval processes within the industry.