Development of a Methodology for the Identification of Ferrous Metal Products by Their Contactless Point Labeling Using Convolutional Neural Networks

The paper proposes a method of contactless matrix labeling of products made of ferrous metals, which combines the simplicity of applying shock-point labeling and the low cost of ink-jet labeling. The main goal of the enclosed solution is to reduce the cost of labeling. Various options for applying labeling at metalworking enterprises are considered. An approach to the recognition of the applied labeling on the surfaces of ferrous metals is proposed. A convolutional neural network was used to develop the recognition system. For training, we synthesized our own dataset of 12,865 images. The trainings of the neural network were carried out, the training results are given. Experimental studies on random images are presented, which showed a high percentage of labeling recognition.