SYSTEM FOR SEGMENTING IMAGES OF BUILDINGS AND STRUCTURES Russian patent published in 2020 - IPC G06N3/02 G06T5/50 

FIELD: physics.

SUBSTANCE: invention relates to computer engineering and can be used for segmentation of images of buildings and structures. System comprises computer facilities structured on the basis of a convolutional neural network, as part of encoding units for a three-channel image and an image of an infrared channel and decoding units intended for segmenting images of buildings and structures of small, medium and large size, processing fragments of four-channel satellite images containing images of buildings and structures with additional separation of "stuck together" objects using a pre-conditioned DWT network and generation at the learning phase of the segmentation algorithm of eight fragments of images using all possible variations of rotation of source fragment by angle π/2, π, 3π/2 and mirror image, with application to each new obtained fragment of operators random turn at a small angle within [-π/12; π/12], scale change within [1; 1.2], shift within 2 % of fragment size and chromatic distortion. Result is a satellite image of the original size with buildings and structures selected thereon.

EFFECT: technical result is faster processing of digital data while reducing amount of computational resources.

1 cl, 5 dwg