MODELS AND METHODS OF INTELLECTUAL INFORMATION TECHNOLOGY OF AUTONOMOUS NAVIGATION FOR COMPACT DRONES
Науковий журнал «Радіоелектроніка, інформатика, управління»
Переглянути архів Інформація| Поле | Співвідношення | |
| ##plugins.schemas.marc.fields.042.name## |
dc |
|
| ##plugins.schemas.marc.fields.245.name## |
MODELS AND METHODS OF INTELLECTUAL INFORMATION TECHNOLOGY OF AUTONOMOUS NAVIGATION FOR COMPACT DRONES |
|
| ##plugins.schemas.marc.fields.720.name## |
Moskalenko, V. V.; Sumy State University, Sumy, Ukraine Moskalenko, A. S.; Sumy State University, Sumy, Ukraine Korobov, A. G.; Sumy State University, Sumy, Ukraine |
|
| ##plugins.schemas.marc.fields.653.name## |
navigation; visual odometry; unmanned aerial vehicle; convolutional neural network; neural gas; information criterion; support vector method. |
|
| ##plugins.schemas.marc.fields.520.name## |
Context of this article topics is that the issues of choosing the optimal in information and cost sense of models and methods of data<br />analysis in autonomous navigation systems under a priori uncertainty, resource and information constraints are not sufficiently investigated<br />and are still not fully resolved.<br />Objective – to increase the efficiency of the autonomous navigation system of a compact drone on the terrain, based on data from visual<br />and inertial sensors, operating in training and exam modes in the conditions of limited computing resources and the volume of the training sample, in the information and value sense.<br />Methods of a research are based on the usage of technology of convolutional neural networks for the formation of a feature representation of visual observations, sparse-coding neural gas algorithms for the training of convolutional filters, the model of support vectors for regression analysis of data, on the principles of mathematical statistics and information theory for constructing and evaluating the functional efficiency of classification decision rules.<br />Results: new models and methods of information intelligent technology of autonomous navigation for compact drones have been developed,<br />allowing the training of the most computationally labor-intensive component of the system, the feature extractor from observation, in<br />unsupervised manner in the process of direct propagation of the signal. In this case, the criterion for choosing the optimal parameters in information and cost sense for model of data analysis is proposed, and it is shown by the results of physical modeling that the validity of the<br />formed decision rules is acceptable for practical use.<br />Conclusions. The architecture of the convolutional neural network and the method of its unsupervised learning for the formation of a<br />feature representation of observations in the autonomous navigation problem based on the algorithm of sparse coding neural gas is proposed. The criterion for choosing data analysis parameters is developed and according to the results of physical modeling, the suitability for practical use of developed algorithms of navigation on unknown terrain is proved. The practical value of the results obtained for unmanned aviation is to form a modern scientific and methodological basis for designing capable of training autonomous navigation systems for compact drones operating in the conditions of resource and information constraints. |
|
| ##plugins.schemas.marc.fields.260.name## |
Zaporizhzhya National Technical University 2018-12-07 16:07:43 |
|
| ##plugins.schemas.marc.fields.856.name## |
application/pdf http://ric.zntu.edu.ua/article/view/149599 |
|
| ##plugins.schemas.marc.fields.786.name## |
Radio Electronics, Computer Science, Control; No 3 (2018): Radio Electronics, Computer Science, Control |
|
| ##plugins.schemas.marc.fields.546.name## |
uk |
|
| ##plugins.schemas.marc.fields.540.name## |
Copyright (c) 2018 V. V. Moskalenko, A. S. Moskalenko, A. G. Korobov |
|