A two-layer deep learning-powered optical character recognition for enhance calibration processes

The advent of digital transformation of metrology has led to significant changes in calibration processes. Digital platforms are created to automate machine and calibration processes. This research aims to develop the OCR prototype by integrating deep learning-powered technology in order to increase productivity of calibration processes. This prototype utilizes deep learning techniques, specifically leveraging advanced CNN, to autonomously recognize and to interpret characters, symbols, and numerical data from measuring instrument displays. The initial stage of character detection involved the use of the YOLOv4-tiny CNN for object detection and subsequently character classification using Darknet-19. This developed OCR system was applied in dimensional calibration laboratories. The length calibration by using laser hologauge shows that it can reduce calibration time by up to 24.70 %. Furthermore, accuracy of data transfer is 100 % in a controlled environment and 96.57 % in a non-controlled environment.