Employees of the Laboratory of Photonics and Optical Information Processing at MEPhI have developed an original neural network for the calculation and synthesis of holographic images. The results of the study are presented in the highly rated journal Optics and Lasers in Engineering.

Computationally intensive iterative methods are usually used to calculate (synthesize) diffraction and hologram optical elements capable of optically forming a 3D scene. MEPhI scientists have proposed a neural network method called 3D-CGH-Net, which provides fast calculation of optical elements. At the same time, it has been experimentally demonstrated that the quality of optically reconstructed 3D scenes from such elements can be higher than for synthesized elements using classical, more resource-intensive algorithms. "The speed of calculating high-quality holograms using "conventional" methods is low," says Dmitry Rymov, an employee of the laboratory. - We have developed a method using a neural network of the original architecture and an extensive structure to account for a large set of sections of a three-dimensional scene in the calculated hologram. The network is trained on samples ranging from tens of thousands to hundreds of thousands of examples. The method has been successfully applied in experiments in the optical formation of three-dimensional scenes, in the implementation of holograms using high-resolution high-speed space-time light modulators of the latest types."

According to the head of the laboratory, Professor Rostislav Starikov, computer synthesis of holograms consists in calculating holograms, which can then be implemented in one way or another, for example, from any material, by printing, or using space-time light modulators (in fact, these are microdisplays); The use of computer-synthesized holograms makes it possible to accurately and quickly form specified light distributions (even those that do not exist in nature)."The use of computer-synthesized holograms is promising in creating three-dimensional visualization tools, for laser control of microparticles, for photostimulation of biological neurons, for 3D printing, for converting and focusing light beams, for building holographic memory systems, and much more," explained Rostislav Starikov..

The technique of computer synthesis of holograms has been developing since the late 60s and is now very advanced. To calculate computer-synthesized holograms, it is necessary to solve the inverse problem (calculate the shape of the diffraction element based on the required distribution of amplitude and phase of light formed by it) there are a number of "classical" methods, they are usually iterative and computationally capacious, the calculation of a hologram can take hours, and this often turns out to be unacceptable in modern practice. "The use of neural networks allows you to calculate or, more precisely, generate a hologram if the network is previously successfully trained. It takes time and large training samples to train, but the trained network generates a hologram very quickly," says the head of the laboratory.

"The latest intelligent methods already make it possible to significantly expand the boundaries of the possibilities of using neural networks in solving not only amateur tasks, but also when using them to solve problems in complex scientific topics," says Pavel Cheremkhin, Associate professor of the Department of Laser Physics at MEPhI National Research University.- So, in just a couple of years, we have achieved such a high quality of hologram calculations using neural network methods that it exceeds some of the capabilities of standard methods that have been developing for decades. Our laboratory has developed a method that synthesizes megapixel holograms of complex three-dimensional scenes in just a fraction of a second. At the same time, a high level of optical reproduction of these 3D scenes from such holograms has also been achieved. And the use of modern space-time light modulators for the implementation makes it possible to form thousands of holograms per second and, accordingly, change or transform the three-dimensional light distribution thousands of times per second."