MTUSI: Neural network to improve image quality

Main article: Neural networks (neural networks)

2025: Neural Network Announcement for Image Processing

MTUSI presented a development in the field of artificial intelligence: a neural network for optimizing image quality, available for widespread use. The development of MTUCI will optimize the accuracy of scientific research and industrial analysis with minimal system requirements. This was announced on October 21, 2025 by the press service of MTUSI.

𝓲

Фото: jcomp на Freepik

As reported, employees of the Department of MKiIT, Faculty of Information Technologies, MTUSI - Doctor of Technical Sciences, Professor Yuri Leokhin and Ph.D., Associate Professor Timur Fathulin - have developed a software solution based on a neural network such as an autoencoder that can optimize the quality of photographs and images for use in everyday life and professional fields.

The development is capable of solving critical tasks relevant to many industries - from space research to road construction. Roadbed images, space photographs, macro and micro images require high detail to conduct accurate analysis and obtain reliable scientific data.

How it works: Autoencoders are neural networks that learn to reconstruct input at the output. The technology allows you to remove noise in images, restore lost details and improve clarity. This type of neural network was chosen after a detailed analysis of existing neural network architectures, which are most often used to solve such problems. In the course of research, metrics (criteria) for assessing image quality were also determined, a unique architecture of the autoencoder was designed and its training was carried out.

The main difference developed by scientists from the MTUSI neural network is the optimal efficiency, expressed in the speed of image processing and low system requirements for end-user hardware. If most similar solutions require powerful hardware, then MTUSI development can work on conventional computers, which makes the technology available to a wide range of users.

Model training was conducted on the Google Colab platform using the Tesla T4 GPU. The system was trained on 53 eras (out of a planned 200), after which it showed optimal results. Several models with different parameters were tested to determine the most efficient configuration.

The creation of neural network technology to improve the quality of images is a step forward in the development of the Russian AI developments. In conditions of technological sovereignty, it is especially important that our scientists create competitive solutions. This development demonstrates that MTUSI not only prepares personnel for, digital economy but is also an active participant in the creation of modern technologies. Of particular value is the practical applicability of our neural network - from space research to industrial quality control. We see the significant potential of this technology for Russian science and, industries especially given its effectiveness, which makes artificial intelligence available to researchers throughout. to the country told Sergey Dmitrievich Erokhin, rector of MTUSI

In the near future, it is planned to adapt the developed software solution for highly specialized branches of science and technology. The technology can find application in:

• Space research - optimizing images from telescopes and satellites
• Road Construction - Roadbed Quality Analysis
• Industrial Inspection - Image Detail for Defect Detection
• Scientific research - working with images of micro- and macromir