ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 6. Vol. 25. 2019

DOI: 10.17587/it.25.368-372

A. V. Zharkov, Deputy Head of the Science and Technology Center, ajarkov@npomis.ru, Scientific and Production Corporation "Mobile Information Systems"

Formation of Classification Signs of Object Recognition when Using Laser-Location Means of their Scanning

The task of effective selection of ground-based objects based on the results of the use of laser means of scanning them can be solved provided that the object is reliably selected against the background of a complex masking relief. The recognition result will be identification (assignment of the proper and unambiguous name to the object in question) or classification (determination of the object's belonging to a given class with an estimate of their coordinates). Identification or classification of objects is made on the basis of an a priori dictionary of attributes and the alphabet of their division into classes. In the self-learning recognition system, the feature dictionary can be filled (corrected). The development of a generalized algorithm for the formation of a feature space required for reliable separation of image classes is one of the most important and time-consuming tasks of creating a system with recognition, since in most cases the distinctive features do not have clear boundaries and cannot always be formalized. Selection of features can be carried out in the measurement space and the transformed space. Reducing the number of features reduces the cost of carrying out measurements and calculations, but can lead to a drop in the reliability of recognition. If the time for training and decision making is severely limited, then increasing the dimensionality of the attribute space may be the only means of increasing the reliability to the required level. Thus, from a practical point of view, the requirements of the minimum of the overall dimension of the problem of recognition and the maximum of confidence are in conflict. In this article, the author examines the task of forming an a priori alphabet of classes, based on an a priori compiled dictionary of object recognition features from the results of an airborne laser radar scan.
Keywords: laser location, relief, feature dictionary, class alphabet, recognition, a posteriori probability, a priori probability, matrix of winnings

P. 368–372

To the contents