EUVIMEDEUVIMED entdecken
Crossref

Architecture and Implementation of a Software System for Wheel Identification by Markings Using Incremental Learning

A. V. Mareev, A. A. Orlov, I. D. Vereshchagin

Programmnaya Ingeneria · 18.05.2026 · 17 (5) · 280-292

DOI: 10.17587/prin.17.280-292 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Programmnaya Ingeneria, erschienen am 18.05.2026. Verfasst von A. V. Mareev und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This article addresses the task of ensuring end-to-end traceability in railway wheel production lines using modern computer vision methods. The aim of the work is to design and develop a software system for the automatic identification of wheels based on their markings, which is a critical element for the digitalization of manufacturing processes. The algorithm proposed by the authors is a multi-stage video processing pipeline based on the YOLOv11 family of neural network architecture and the ByteTrack object tracker. The pipeline involves the sequential execution of the following operations: detection and tracking of wheels in frames, precise localization of the marking area, its geometric alignment and cropping, segmentation and classification of individual characters, and the formation of a final marking based on a consensus approach that aggregates results across a series of frames. A key feature of the system, distinguishing it from traditional solutions, is the implementation of an incremental fine-tuning module. This module allows for continuous improvement of recognition accuracy during operation by adapting the model to new marking types without the need for complete and resource-intensive retraining on the entire dataset. The article presents the results of a comprehensive experimental study confirming the system's high efficiency. A detection performance of mAP@0.5 = 0.995 was achieved on test data, with the system demonstrating robustness to typical industrial noise (contamination, glare, lighting variations). The system's performance reaches up to 35 frames per second on high-performance hardware. The modular architecture of the developed standalone application and its user interface are described in detail. The interface ensures convenient operator interaction with the system, including capabilities for manual correction and initiating the incremental fine-tuning process. The practical significance of the work lies in creating a universally applicable tool, robust to industrial noise, applicable not only in the railway industry but also in other fields of mechanical engineering for solving automated identification tasks for large-scale products.

Zitieren

APA

Mareev, A. V., Orlov, A. A., & Vereshchagin, I. D. (2026). Architecture and Implementation of a Software System for Wheel Identification by Markings Using Incremental Learning. Programmnaya Ingeneria, 17(5), 280-292. https://doi.org/10.17587/prin.17.280-292

Vancouver

Mareev AV, Orlov AA, Vereshchagin ID. Architecture and Implementation of a Software System for Wheel Identification by Markings Using Incremental Learning. Programmnaya Ingeneria. 2026;17(5):280-292. doi:10.17587/prin.17.280-292

BibTeX
@article{Mareev2026,
  author = {Mareev, A. V. and Orlov, A. A. and Vereshchagin, I. D.},
  title = {Architecture and Implementation of a Software System for Wheel Identification by Markings Using Incremental Learning},
  journal = {Programmnaya Ingeneria},
  year = {2026},
  volume = {17},
  number = {5},
  pages = {280--292},
  doi = {10.17587/prin.17.280-292}
}
Kostenlos starten →