Decoding Dot Peen Data Matrix Code with Deep Learning Capability for Product Traceability

Abstract

An approach for recognizing and decoding the industrial-based dot peen data matrix code is presented in this paper. Dot peen marking is a type of direct part marking (DPM). Due to the reduced contrast characteristic, it could be difficult to read a DPM code. Additionally, the readability of a DPM code may deteriorate over time due to partial degradation on the product surface. A deep-learning-based method using You-Only-Look-Once (YOLO) v5 model is proposed. Firstly, a large dataset of dot peen data matrix symbols was prepared to initiate the YOLOv5 model training. Image data augmentation was then applied to the training images to increase the size of the training dataset. The YOLOv5 model training was processed with a batch size of 16 and the epochs number of 60 due to its high accuracy (97.79%). All dot peen data matrix codes were detected accurately within one second, fulfilling our intention to design a high-speed reader for industrial-based dot peen data matrix. With ANOVA analysis, we observed that the brightness level and the camera distance significantly affect the decoding process. Additionally, our developed model can successfully decode a partially damage code if the level of damage is below 30%.

Keywords

• data matrix code
• deep learning
• direct part marking
• dot peen marking