MTUSI: Neural network for warehouse accounting automation

Main article: Neural networks (neural networks)

2025: Announcement of a neural network for warehouse automation

Scientists of the Moscow Technical University of Communications and Informatics (MTUSI) have developed a system based on artificial intelligence, which automatically determines the number and type of pipes in warehouses and production sites. The technology is able to replace the manual labor of workers and optimize warehouse accounting processes in the metallurgical industry. This was announced on October 21, 2025 by the press service of MTUSI.

𝓲

Фото: Freepik

As reported, employees of the Department of MKiIT Faculty Information Technology"" MTUSI - Doctor of Technical Sciences, Professor Yuri Leokhin and Ph.D., Associate Professor Timur Fathulin - created a software solution based on neural network YOLOv8, specially adapted for recognitions metal products real production conditions.

Despite the abundance of software products for automating business processes, many specific operations at Russian enterprises are still carried out manually. In production companies of metal structures, the calculation of pipes after unloading and before the start of production is traditionally carried out by an employee without any automated means - a process that is time-consuming, long and prone to errors.

The development of MTUSI solves this problem comprehensively: the computer vision system is able to recognize pipes in photographs taken by a regular smartphone, determine their type and automatically count the number. This is especially important for enterprises operating with large volumes of metal products.

Technological capabilities:

• Real-world training: The neural network was trained on a datacet of 5,800 images, including both open data Yandex"" and real-world photographs from manufacturing facilities. A key feature was the data augmentation technology - the generation of additional images with different effects to simulate different shooting conditions.
• Working in difficult conditions: The developers paid special attention to the analysis of the influence of external factors on the accuracy of recognition. The system has been tested at different viewing angles, at different degrees of illumination and for different types of pipes. Based on the results of experiments, recommendations were formulated for the optimal application of the algorithm in real production conditions.
• Optimal accuracy: The main possibility of development is the ability to accurately recognize objects even in photographs of low quality taken on conventional smartphones in production conditions.

This development clearly demonstrates how artificial intelligence can solve specific problems of the real sector of the economy. We have created a tool that will allow Russian metallurgical enterprises to optimize operational efficiency and reduce operating costs. It is especially important that our neural network works with photos taken on ordinary smartphones - this makes the introduction of technology simple and affordable. There is no need to buy expensive equipment, enough phone camera and our software. Such developments confirm that MTUSI not only trains highly qualified specialists in the field of IT and artificial intelligence, but also creates applied solutions that can be immediately introduced into production. This is a contribution to the digitalization of Russian industry and increasing its competitiveness. told Sergey Dmitrievich Erokhin, rector of MTUSI

The introduction of an automated pipe definition system will allow enterprises to:

• Reduce inventory time - from hours to minutes
• Eliminate the human factor - eliminate manual counting errors
• Optimize warehouse processes - get real-time data
• Reduce labor costs - free up staff for more complex tasks
• Improve accounting accuracy - ensure transparency of inventory balances

The developed system can be adapted to the specific needs of various warehouses and production facilities. The technology is applicable not only for pipes, but also for recognizing other types of metal products and building materials.

Target industries for implementation:

• Steel Mills
• Steel structure manufacturers
• Construction companies
• Logistics warehouses
• Pipe rolling plants

Specifications:

• Architecture: Neural Network YOLOv8
• Training Sample Size: 5,800 Images
• Data sources: Yandex open resources + production photos
• Working conditions: photographing with a smartphone
• Adaptability: working at different angles, lighting, and pipe types

The creation of specialized AI solutions for Russian industry is a strategically important area in conditions of technological sovereignty.