APPARATUS FOR SELECTING THE CONTOURS OF OBJECTS IN HYPERSPECTRAL IMAGES Russian patent published in 2022 - IPC G06K9/62 

FIELD: image data processing.

SUBSTANCE: invention relates to the field of digital processing of hyperspectral images, in particular, to selecting the contours of objects in hyperspectral images. The technical result is achieved by additionally introducing, in the apparatus for selecting the contours of objects in hyperspectral images consisting of L gradient image generation blocks, the inputs whereof are the input of the apparatus, the adder, and the threshold comparison block, the output whereof is the output of the apparatus and the second input is a process input, a block for forming K combinations of paired non-repeating combinations of gradient images from L gradient images, said block having K paired outputs, wherein L is the number of gradient images, K is the number of paired combinations, ! is the factorial; and K blocks for calculating the difference modules of like pixels of each kth pair of gradient images, wherein k=1,…,K; a divider, the output whereof is connected with the input of the threshold comparison block and the first input is connected with the output of the adder; a counter, the output whereof is connected with the second input of the divider, wherein the output of the gradient image generation block, where is connected with the corresponding input of the block for forming K combinations of paired non-repeating combinations of gradient images from L gradient images, the kth paired output whereof is connected with the first and second inputs of the corresponding kth block for calculating the difference modules of like pixels of each pair of gradient images, the kth output whereof is connected in parallel with the corresponding kth input of the adder and with the corresponding kth input of the counter.

EFFECT: higher accuracy of selecting the contours of spectral-selective objects by considering the additional interrelations between the spectral components of the gradients of the hyperspectral image.

1 cl, 9 dwg