ISSN 0236-235X (P)
ISSN 2311-2735 (E)


Next issue

Publication date:
16 September 2020

Journal articles №2 2020

21. Setting up and training a multilayer perceptron for the problem of highlighting the road surface in the city space images [№2 за 2020 год]
Authors: V.S. Tormozov ( - Pacific National University (Senior Lecturer); K.A. Vasilenko ( - College of Service and Design at Vladivostok University of Economics and Service (VGUES) (Lecturer); A.L. Zolkin ( - Samara branch "Volga State University of the Water Transport" (Lecturer), Ph.D;
Abstract: The paper considers the neural network model application for a multi-layer perceptron to identifying road surface region problems on the urban environment satellite images. Government agencies and en-terprises involved in regulating transport flows currently need to solve this problem, as well as to up-date geographical data and maps of transport infrastructure. In existing works on this topic, there are automatic and semi-automatic methods. Approaches that involve a person’s partial involvement in their work are semi-automatic. The operator can set thresholds, setting parameters, mark regions for detection, and do many other operations. Automatic methods work without human involvement and therefore faster and cheaper than semi-automatic ones. The paper examines and explores a method that uses a multilayer neural network to automatically highlight the road surface on the Earth 's surface cosmic images. In its paper, the method is based on a limited sample of roadway previously noted examples. The model has a multilayer perceptron founda-tion. The input values for the method in question are satellite survey data in the RGB color model. This makes it possible to use more information channels individually. This also takes into account the pixel channel context: values of the image adjacent pixel color channels. The method in question is relevant, as the expansion of street road network and urban development are changing and should be reflected in the mapping data. The results of the method were with the lo-cation of the road surface of the city's road network.
Keywords: digital image processing, artificial intelligence, machine learning, pattern recognition, roadway detection, street road network, artificial network, satellite imagery
Visitors: 299

22. Comparative analysis of community identification algorithms in complex network systems using social networks as an example [№2 за 2020 год]
Authors: Kochkarov A.A. ( - OJSC "RTI", Financial University under the Government of the Russian Federation (Deputy Director R&D centre), Ph.D; N.V. Kalashnikov ( - Finance University under the Government of the Russian Federation (Applicant); R.A. Kochkarov ( - Finance University under the Government of the Russian Federation (Associate Professor of Department of Data Analysis, Decision Making and Financial Technologies), Ph.D;
Abstract: The paper considers the identifying community in social networks. There is a graphical approach to the study of social networks. There is a comparative analysis of the basic algorithms and the aggregate al-gorithm proposed by the authors. To test the algorithms, the authors generated graphs initially with different noise levels and gave a community number. To compare graph partitions, two well-known metrics the authors used – Normal-ized Mutual Information (NMI) and Split join distance. Each of the metrics has its own advantages. To verify the basic algorithms, and analysis the authors made of the Facebook social network geo-graphic for the community presence in them and tested the aggregated MetaClust algorithm. The pro-posed MetaClust algorithm showed high performance compared to the base ones. The modularity val-ues for its partitions (on average) are higher compared to the basic algorithms. Also, the algorithm qual-ity can be judged by the absence of a “tail” modularity in the distribution. The average results shown by the algorithms on the generated graphs correspond to the application results on the ego networks. To generate model data, it seems appropriate to use pre-fractal graphs and a wider class of dynamic graphs. The sequence of generated community graphs corresponds to the dynamic graph trajectory, the communities are seeds and blocks, and the noise is the addition of the new edge different ranks be-tween the seeds. The next step is a formal description of the graphs’ noise in the class terminology of the dynamic and pre fractal graphs. Using the pre fractal graph class will allow us to calculate the structural charac-teristics and of graph properties and communities in them.
Keywords: dynamic graph, aggregated algorithm, basic algorithms, communities, social networks
Visitors: 284

← Preview | 1 | 2 | 3