A new Russian technology will allow to replace a human in one more area
The scientists from the National Research Nuclear University MEPhI (MEPhI) have managed to be the first to train the neural network to retrieve and analyze the nanoparticles with a microscope. The method of the neural network training proposed by them enables to eliminate the manual processing of microphotographs, which greatly increase the speed and quality in the analysis of new nanomaterials.
Usually both the analysis of microphotographs and their labeling for neural networks is carried out manually: the scientist circles each particle and measures its dimensions. In this case, there can be thousands of them in one picture. Existing neural network architectures allow to perform the efficient analysis of such images, so the problem, according to the scientists, is only in unavailability of a sufficient array of labeled data.
It turned out, according to the scientists of MEPhI, it is possible simply to draw the nanoparticles in view of their texture in real microphotographs from the open databases of images made by a scanning electron microscope (SEM). Then, it will be known exactly, where the particles are located and what are their dimensions.
Еру scientists believe that the results of the study will allow to automate the SEM image processing, thus, revolutionizing the standard methods of research of new materials. This will not only enable to reduce the durability of the study, but also increase the number of the analyzed particles from hundreds to tens of thousands.
In the future, the scientific team intends to use the same approach to classify the nanoparticles by the shape, insofar, now it is also the entirely manual work.
Source: RIA Novosti