AN APPROACH TO FORMING A PRIOR DATASET OF NEURAL NETWORK FOR PROCESSING DIGITAL AERIAL PHOTOS

Analysis of the use unmanned aerial systems in combat zone indicates a constant increase requirements for intelligence data. One ways to effectiveness UaV application is artificial methods. The basis building reliable neural network model large amount various data, which helps more accurately summarize information about given task. article considers an approach increasing accuracy object image classification by convolutional networks based on data augmentation, differs from existing ones its adaptation shooting factors and specificity reconnaissance objects. An important most time-consuming step accurate machine learning finding annotating that will be used train test network. stability real conditions depends collected To date, such methods augmentation as geometric transformations, color correction, spatial filtering have been well described analyzed. possibilities optimal combinations achieve desired generalization poorly visible invariant features objects remain unexplored at moment. Taking into account conducted research, was proposed formation set priori recognition digital photographs, significantly reduce complexity process collecting necessary replace it with magnification are much simpler, consume less computing resources, work networks, simulation carried out. Keywords: vehicle; network; augmentation; automated processing photographs.