ALGORITHMS OF CLUSTERING AT DECODING PHOTOES OF UNMANNED AERIAL VEHICLES (DRONES)
Tatyana Karlova, Tatyana Karlova, Гурьев Александр Тимофеевич, Гурьев Александр Тимофеевич, Роман Алешко, Роман Алешко, Ксения Шошина, Kseniya Shoshina, Сергей Шептунов, Sergey Aleksandrovich Sheptunov · Bulletin of Bryansk state technical university · 2017
The procedures and algorithms automating a process of a multilevel subject decoding are of particular interest. In the paper there is described a development of algorithms for automatic object identification on the basis of clustering. In the investigation the algorithm for a cluster analysis of AKM (improved of k- means) which allows identifying first an object in the picture and then highlighting it graphically is used. This algorithm is formed on the basis of the k-means al-gorithm allowing the fulfillment of a rapid cluster analysis. The improvement of AKM algorithm consists in a possibility of the computation of an optimum cluster number at a specified maximum cluster number. The accuracy of the results of subject decoding is as-sessed. One of the methods for the assessment of relia-bility is a statistic assessment of picture decoding re-liability. For this it is necessary to create a matrix of errors at cluster definition and to calculate accuracy. It is possible to use a method of cross-tabulation for the presentation of pixels defined correctly in an obtained subject map of forest roads and a map formed on the basis of UAV pictures and data of ground investiga-tions. A general accuracy of object decoding as a re-sult of the work of AKM algorithm and a procedure of UAV picture processing of forest roads characterizes a degree of reliability as a high one. The options for the further improvement of a procedure and algorithms are offered.