+7 (495) 987 43 74 ext. 3304
Join us -              
Рус   |   Eng

articles

Authors: Magomedova S., Feylamazova S., Millinov M.     Published in № 4(106) 25 august 2023 year
Rubric: Researching of processes and systems

Intelligent system of bulk raw material quality assessment based on YOLOv5 and ViT neural networks

The lack of an effective system of raw material quality assessment during acceptance in the field is the reason for writing this article. The article describes important scientific and practical tasks related to this problem, such as the development of methods of raw material quality control, creation of algorithms of intellectual support for fast and accurate assessment of raw material quality in real time. The authors have analyzed domestic and foreign scientific papers describing approaches and methods for raw material quality control and concluded that the YOLOv5 algorithm and visual transformer are the most suitable for the present study, their combination has not been previously used in the literature. The paper presents the problem statement for the development of an intelligent raw material quality system, description of object detection algorithms, their comparative characterization. The description of the working principle of the one-stage algorithm of object detection in the image YOLOv5 and its advantage over the existing analogs of deep learning methods, which are widely used for object detection, is given. Also presented are the results of neural networks, the task of which is to identify defective raw materials from the total amount stored in a particular railcar and belonging to a particular supplier. An intelligent system for analyzing the quality of bulk raw materials includes cameras located above and to the side of railcars. The quality of raw materials is determined by transferring images from the loading cameras to a neural network for further processing. YOLOv5 algorithm is used to detect the railcar number, which provides accurate detection of railcar number. To determine the percentage of rejected cargo, in the first step, YOLOv5 approach is used to select the image region with cargo and then visualization is applied to detect the percentage of rejected cargo. Implementation of the proposed intelligent system will allow more efficient production management, reduce costs and improve the quality of delivered bulk raw materials.

Key words

quality control system, raw materials, YOLO, computer vision, neural networks

The author:

Magomedova S.

Degree:

Cand. Sci. (Econ.), Associate Professor of Information Systems and Programming Technologies Department, Director of Information and Computing Center, Dagestan State University

Location:

Makhachkala, Russia

The author:

Feylamazova S.

Degree:

Senior Lecturer, Information Technology and Computer Systems Security Department, Dagestan State University

Location:

Makhachkala, Russia

The author:

Millinov M.

Degree:

Master’s Student, Information Systems and Programming Technologies Department, Dagestan State University

Location:

Makhachkala, Russia