AUTOMATION OF BARCODE READING IN ERP SYSTEMS

Abstract

The article considers the issue of automating material flow management processes in ERP systems using the example of welding electrode production. Particular attention is paid to the implementation of barcode technologies to improve inventory management efficiency, reduce errors due to the human factor, and accelerate production processes. An analysis of modern barcode reading devices, such as hand-held scanners, smartphones, and conveyor cameras, is conducted to determine optimal solutions depending on the application conditions.

The main goal of the study is to automate material accounting through the development of functional modules in the PlazmIS ERP system, which is used by the PlasmaTek company. The work proposes a module called "Batch-Container Tag" that allows you to generate and print labels for marking raw materials, as well as models for the barcode detection module.

A two-stage approach is proposed to automate barcode reading using smartphones, which includes the stages of barcode detection and recognition. The article considers the first stage - barcode detection. The ParcelBar dataset containing 844 barcode tag images was used for training. The best model based on the YOLOv8 small architecture according to the mAP@(0.50-0.95) criterion achieved a level of 0.930 for barcode detection, which is 1.2% higher than the result of the model based on the YOLOv5 small architecture and 7.5% higher than EfficientDet_0. In the course of the research, an updated ParcelBarRelabeled dataset was also created with a new markup, where the bounding boxes do not cover the entire tag, but only the barcode area. The best YOLOv8 nano model according to the mAP@(0.50-0.95) criterion achieved a level of 0.835 for barcode detection on the validation data of this dataset. To improve accuracy on specific data of products of the enterprise "PlasmaTek" the dataset ParcelBarRelabeledExtended was created, which allowed to reduce the number of false detections and improve the level of confidence of detections on images with raw materials of the enterprise. The best model YOLOv8 nano according to the criterion mAP@(0.50-0.95) reached the level of 0.86 on the validation data of this dataset.

The research results indicate the potential of using YOLOv8 for mobile devices due to its high accuracy and speed, but the need for further optimization of the models is noted.