RU 
EN 
CN 
ES 
Scientists from Samara University and ITMO University have developed the innovative method for detecting wave distortions that reduce image quality. According to the new approach, light passes through a spatial optical filter, a device that gives way to or blocks certain parts of the light beam, depending on their direction or form. This results in shaping a complex image containing information about wavefront distortions. The neural network then analyzes the resulting images.
The proposed method made it possible to achieve 99.7% accuracy in recognizing wavefront distortions in real time. This opens up new opportunities for astronomy, ophthalmology, and high-precision metrology. The research results, which has been supported by a grant from the Russian Science Foundation (RSF), are published in the scientific journal “Technologies”.
Wave aberrations are distortions of the wavefront, that is, the surface on which all points of the light wave have the same oscillation phase at a given time. These distortions occur if light passes through the turbulent atmosphere, defective optical elements, and unbalanced optical systems such as telescopes, microscopes, or even a human eye. Wave aberrations lead to decreasing image clarity, so scientists are looking for ways of dealing with them.
Researchers from Samara National Research University and ITMO University have proposed using a hybrid multichannel diffraction optical element, i.e. a complex optical element with a microstructure on its surface, for detecting wave aberrations. It is implemented on the base of a spatial light modulator, a device that allows adjusting the light-beam phase distribution (“phase” stands for “position” of the wave at a specific moment). This optical element transforms a distorted wavefront into a visual multi-channel image that can be used for determining the type and the extent of the wavefront distortion.
The neural network analyzes the images obtained by using the optical filter in real time and recognizes aberrations. The researchers used the neural network based on the Xception architecture, which is often used for classifying and segmenting images, as well as detecting objects on them. The algorithm was trained on 2,352 images, each 256×256 pixel. The error in recognizing wave aberrations was only 0.3%.
The developed high-precision method can be used in astronomy for adjusting atmospheric distortions in telescopes, in ophthalmology – for early diagnosis of eye diseases, in industry – for quality control of lenses and mirrors, as well as in quantum metrology.
“This study demonstrates how combining modern optical technologies with machine learning makes it possible to solve complex problems with unprecedented accuracy. The method has already revealed its efficiency in the laboratory, and its introduction in industrial and medical applications is a matter of the near future. In future, we plan to develop a small-sized wavefront sensor with software and hardware based on a hybrid multichannel diffraction optical element, for the purpose of fast and accurate measurement of wavefront distortions”, explained Pavel Khorin, the leader of the project supported by the RSF grant, Candidate of Physics and Mathematics, Senior Researcher at Samara National Research University’s Laboratory of Automated Research Systems.
Source: RSF Press Service
