The work was partially financially supported by RFBR in the context of scientific projects N o N o 17-29-03370 and 19-29-09075.

No abstract

In this paper, we propose an approach to the classification of high-resolution hyperspectral images in the applied problem of identification of vegetation types. A modified spectral-spatial convolutional neural network with compensation for illumination variations is used as a classifier. For generating a training dataset, an algorithm based on an adaptive vegetation index is proposed. The effectiveness of the proposed approach is shown on the basis of survey data of agricultural lands obtained from a compact hyperspectral camera developed in-house.

A thorough analysis of the literature on retinal image stabilization, as well as our own experimental data, present evidence that Yarbus's concept, implying inevitable and irreversible fading of a visible image evoked by stabilized retinal stimulus of any size, color, and luminance in 1 to 3 s after its onset, is not valid in a general case. It has been demonstrated that, even with Yarbus's stabilization techniques, the lifetime of visible images varies from fractions of a second to the whole stimulus duration-up to 30 min in our experiments-depending on many factors: monocular or binocular viewing, stimulus parameters, characteristics of subjects, and so forth. The dynamics of perceived images is determined mainly by the processes at the higher levels of the visual system. In the cases of such unusual visual stimuli as stabilized retinal images, it is problematic for the visual brain to find their proper interpretations in terms of everyday natural experience. Usually, the responses of retinal units are determined by three types of coexisting images: (a) the optical projections of external objects, (b) shadows of the blood vessels and other internal eye structures, (c) virtual patterns caused by the traces of previous stimuli. A task of the visual system is to recognize and visualize only external objects separating their projections from all the entoptic images of the two remaining types. To implement separation, visual brain employs a number of approaches--in particular, the eye movements that cause sliding over the retina but only the projection of the external objects. This means that the peculiar phenomena observed in the cases of stabilized retinal images can be determined not by invariability of such stimuli per se but rather by the fact that stabilization eliminates a powerful cue helping to identify the retinal images belonging to the external objects, thereby increasing the probability to treat them as the entoptic ones which should be ignored or canceled rather than perceived. However, the probability of canceling--image fading--can be essentially reduced in conditions of concordant, large, bright, and sharp binocular stimuli.

No abstract