Spotting Danger in a Single Millimetre: BelSU’s Hybrid Neural Network Sees What Human Eyes Miss

Scientists at Belgorod National Research University (BelSU) have created a hybrid neural network capable of automatically identifying dangerous cracks and defects in building structures from photos and videos – tasks that would take human experts dozens of times longer to complete.

The development addresses a critical challenge in construction safety.

“If a crack in a wall or foundation is less than one millimetre wide, not every expert will be able to recognize and document it. This is essential for the timely prevention of potentially serious accidents,” explains Konstantin Polshchikov, Professor at the Department of Information and Robotic Systems and research supervisor.

Yaroslav Golovko, a postgraduate student and co-author of the development, notes that the new models differentiate dangerous defects from visual noise more reliably than existing systems. The technology has particular significance for detecting damage caused by destructive impacts on buildings.

The team plans to expand the system’s capabilities soon, adding detection of dangerous chips, delaminations, and rusted areas. Rector Evgeniya Karlovskaya highlighted the achievement on her social media channel, calling it confirmation that BelSU research is meeting the real needs of the times.