Post sponsored by NewzEngine.com

MIL OSI Translation. Region: Russian Federation –

Source: Novosibirsk State University – Novosibirsk State University –

iOk platform (http://yok.nsu.ru/) includes a set of cloud-based digital services for automatic image analysis using deep machine learning and artificial intelligence methods. The platform hosts three services. Universal telegram service No Code ML (https://t.te/nsu_ml_here) is intended for classical training of a neural network on user datasets. Another service is DLgram (https://t.te/nanopartiytsles_nsk) is designed to recognize numerous homogeneous objects of different nature. Includes training of a neural network by the user using a marked area from the same image. Online service ParticlesNN (http://particlesnn.nsu.ru/) is designed for automatic recognition of nanoparticles in scanning probe microscopy (SPM) and electron microscopy (EM) images by a trained neural network.

All services provide for training a neural network on user objects, automatic recognition of objects in images, and the ability to correct recognition results by the user. They also analyze detected objects and determine their parameters, such as quantity, size, area and concentration. The services are able to work with a variety of images—images from electron microscopes, photographs from digital cameras (including from smartphones), and videos. They recognize various objects: nanoparticles, microorganisms, cells, plant seeds, as well as larger objects – animals, plants, various parts, vehicles and much more. At the same time, to work with services, the user is not required to have any special programming skills or understand neural networks. There is no need for pre-processing of the image. The results are provided in the form of information about all detected objects, and the user can correct them if necessary.

— The idea of creating the first online service ParticlesNN came to us from our teacher from the Institute of Catalysis named after. G.K. Boreskov SB RAS Anna Nartova in 2019. Two goals were pursued: to relieve scientists from routine work and save their time. We started with scanning probe microscopy images, which are currently the gold standard for studying and creating new materials. In this case, quite often the task is to characterize images obtained from a microscope: it is necessary to determine, for example, the average size of objects or their number. Scientists had to carry out these manipulations manually, spending a lot of effort and time. There were automatic image processing methods based on so-called threshold approaches, but they gave good results only on high-quality images, and noise and highlights were perceived as separate objects, and the results were unreliable. In creating our services, we decided to use modern computer vision methods based on artificial intelligence methods,” said the head of the laboratory of deep machine learning in physical methods Institute of Intelligent Robotics NSU Andrey Matveev.

The scientists followed the standard path for deep machine learning – they marked more than 5 thousand objects and trained the Cascade Mask-RCNN neural network on them, running on the server of the Institute of Intelligent Robotics of NSU. The program produced some errors, but overall the results were quite good: there were errors in the number of objects, but the average size of objects was determined quite accurately. The approach was extended to the analysis of electron microscopy data, the most common family of methods in modern materials science. The results of the work were published in scientific journals and received positive reviews.

— We decided to make this neural network available to users of other laboratories and scientific institutes and created the ParticlesNN web service. The user can upload his image, receive statistical results of its processing and correct them. But this service also has disadvantages – it can only work with those types of objects on which the neural network was trained. We realized that each time training it to work with new types of objects is a rather labor-intensive task, so we decided to develop a service that would allow the user to train the neural network on the objects he needs. This is how the online service DLgram arose, and soon No Code ML. Now, for the convenience of users, we have combined them on one iOk platform. Already, the total number of its users is more than 500 specialists,” explained Andrey Matveev.

Note; This information is raw content directly from the source of the information. This is exactly what the source states and does not reflect the position of MIL-OSI or its clients.

Please note; This information is raw content directly from the information source. It is accurate to what the source is stating and does not reflect the position of MIL-OSI or its clients.

EDITOR’S NOTE: This article is a translation. Apologies should the grammar and or sentence structure not be perfect.

MIL OSI News (multilanguage service)