Analysis of the suitability of open databases of annotated images for deep learning of neural networks in the tasks of information-psychological security

The article presents a methodology for analyzing the degree of suitability of open databases of annotated images for deep learning of neural networks in order to ensure information and psychological security of Internet users. The study of the correspondence between the model of representation and recognition of visual images by convolutional neural networks and the peculiarities of visual perception of these images by humans has been carried out. A comparative analysis of the results of experimental studies on the visualization of maps of signs of the trained convolutional network and points of fixation of the gaze of the subjects for the same visual stimuli is carried out. Approaches are proposed to improve the quality of the results of such studies in the future.