.
Development of theoretical and technological fundamentals of artificial intelligence, big data preprocessing and analysis methods and models, machine learning and semantic data systems methods and models. Pattern recognition and classification. Technologies for automated detection and classification of ground and overwater objects using statistical and neural network algorithms.
.

COMPUTER VISION:

.

Competence:

Data preprocessing, machine learning, deep learning, convolutional neural networks, recursive neural networks, auto-encoders, encoder-decoder, nuclear image transformations, assemblies of deep learning models.
 
Used tools – Frameworks, libraries, languages:
 
Opencv 3 и 4, python 3, c++, CUDA, opencl, keras, tensorflow, pytorch, Theano, albumentations, catalyst, mmdetection.
.

The implemented solutions:

.

• Assemblies of two-dimensional convolutional neural networks for the analysis of lung pathologies on CT images
• Construction of a classifier of images of pathological formations obtained using video endoscopy using deep learning methods.
• Detection and recognition of advertising banners on images
• Face detection in high resolution video stream
• Tracking (detection + re-identification + tracking); palm segmentation, key point search.
• Tracking people by fisheye camera.

.

Areas under consideration:

.

• Diagnosis of pigmented skin lesions.
• Gesture recognition algorithms on video sequences
• Neural network face detection algorithms in ultra-high resolution video streams
• Gabor filters in the face recognition task to form a feature vector
•  Each of the deep learning tools that we use is capable of working on GPUs from NVIDIA. The choice between the high-level language Python 3 and the low-level language C ++ should be determined on the basis of the task, namely the parameters of real time operations, and the amount of information analyzed per unit time.

.

Cases:

.

• Module for recognition of persons with disabilities:

To evaluate this problem, we need an example of a data set, using which it is planned to perform recognition, after receiving a data set and studying it, we can give feedback regarding the prospects for solving this problem.

• Performance-efficient vehicle number recognition algorithm:

The problem to be solved, more precise technical details are needed, namely, an example of the hardware configuration, using which the solution of this problem is planned.

• Model for classification of vehicle types:

An example dataset is needed for detailed study. However, one of the hypotheses is the solution to this problem using convolutional neural networks in the configuration of an encoder - decoder. Similar kinds of tasks are successfully solved by neural networks of this type.
.

ANOMALY DETECTION IN INDUSTRIAL TIME SERIES:

The goals are:

.

• develop anomaly detection models based on machine learning algorithms;
• develop software compatible with SCADA-systems for prediction equipment failure.
• Expected Outcomes and Results:
• reliable anomaly detection model for single dimension non-stationary and dynamic time series with adaptive ability, model for multi-dimension time series;
• a novel methodology for prognosis failure on production based on anomaly detection, for pre-processing and feature extraction from industrial series;
• a novel methodology for modeling different type anomalies in industrial time series;
• developing software for predictive analytics on industrial production.
• Current state of the problem:
• obtained models for anomaly detection in single dimension time series. Developed MVP for anomaly detection with ability connect to SQL-database and CSV files.

.

Project implemented by the group:

.
The megaproject “Creation of a domestic high-tech software and instrumental complex for the implementation of process control systems based on free software” (carried out within the framework of the decree of The Government of the Russian Federation No. 218 of 9.04.2010 in 2016 – 2018).
.

Recent significant papers:

.

1. Zamyatin A.V., Afanasyev A.A., Cabral P. Approach to the Analysis of Land Cover Dynamics, using Change Detection and Spatial Stochastic Matelling //Optoelectronics, Instrumentation and Data Processing. 2015. Vol. 51, № 4. P. 354-363.
2. Padmanaban R., Bhowmik A.K., Cabral P., Zamyatin A., Almegdadi O., Wang S. Modelling Urban Sprawl Using Remotely Sensed Data: A Case Study of Chennai City, Tamilnadu //Entropy. 2017. Vol. 19, № 4. P. 1-14.
3. Afanasyev A., Zamyatin A. Hybrid Landscape Change Detection Methods in a Noisy Data Environment //Lecture Notes in Electrical Engineering. 2019. Vol. 520. P. 71-78.
4. Gavrin S., Murzagulov D., Zamyatin A. Detection of Change Point in Process Signals by Cascade Classification //2018 International Russian Automation Conference (RusAutoCon 2018), Sochi, 9-16 september 2018. Vol. 1-2. New York: IEEE, 2018. P. 515-518.
5. Murzagulov D.A., Zamyatin A.V., Ostrast P.M. Approach to Detection of Anomalies of Process Signals Using Classification and Wavelet Transforms //2018 International Russian Automation Conference (RusAutoCon 2018), Sochi, 9-16 september 2018. Vol. 1-2. New York: IEEE, 2018.

.

VIRTUAL UNIVERSITY 4.0.
The goals are:

.

• to create an open IT platform for the development of high-tech scientific and educational products in a wide range of subject areas based on VR/AR technologies;
• to form a new cultural and educational contour of the region and the country, to concentrate the university’s efforts on the production of educational know-how, to complement the university’s innovation ecosystem with a powerful tool to improve the quality of education by creating innovative technologies..

.

Expected Outcomes and Results:

.

• the implementation of scientific and applied projects for the tasks of personalized medicine, the entertainment industry, the creation of production simulators for industry 4.0, elements of the digital economy and their testing;/

.

Current state of the problem:

.

• The scientific laboratory of VR/AR has been created at TSU, which is equipped with advanced equipment. In the laboratory, students can work with interactive 3D-models of equipment and practice the mechanics of important processes, and teachers can create interactive courses based on a library of models and tools. Immersive technologies and products will be used for scientific and industrial purposes

.

Project implemented by the group:

.
The project “Scientific and methodological foundations of the construction of software and hardware systems of multidimensional visualization for solving problems of monitoring and controlling of infrastructure facilities”, No. 2.4218.2017 / 4.6). (performed in the framework of the State task "Science" in 2017 – 2019)
.

THE DIGITAL PLATFORM FOR THE MEDICAL DATA ANALYSIS

 

The goals are:

.

• creation of a single complex with broad range of instruments  for solving different tasks of various medical data processing;
• increasing the availability of data analysis methods for healthcare providers.

.

Expected Outcomes and Results:

.

• algorithms for analyzing data of various nature and their dynamics using change fixation were developed and tested on model and real data simultaneous data;
• designed and implemented a software and information platform that combines the basic functionality for collecting, storing and processing data while maintaining the ability to scale the system and use relevant and promising diagnostic data formats;
• implementation of the developed intellectual platform will reduce the time of diagnosis, which will      subsequently reduce the burden on specialists, as well as increase the efficiency of early diagnosis of  diseases due to effective algorithms for detecting changes, and therefore the duration and quality of life of people.

.

Current state of the problem:

.

• The software and information platform was prepared as a commercial product, including marketing research, monetization mechanisms, to bring it to the domestic and world markets.