METHOD FOR MULTISPECTRAL RECONSTRUCTION OF NIGHT IMAGES OF URBAN AGGLOMERATIONS Russian patent published in 2025 - IPC G06T1/40 

FIELD: data processing.

SUBSTANCE: invention relates to digital image processing and relates to a method for multispectral reconstruction of night images of urban agglomerations. When implementing the method, for each pixel of the image of the panchromatic satellite image, an area corresponding to this pixel is determined on a high-resolution map of built-up areas and the average building coefficient and its standard deviation are calculated. Pseudo-multispectral image is formed from this area using the brightness values of the pixels of the panchromatic image instead of the red layer, the average building factor instead of the green layer and the standard deviation of the building factor instead of the blue layer. Image is cut into fragments. Fragments of the image are fed to the input of three neural networks which encode one of the three colours of the RGB model. At the output of each neural network, a representation of the satellite image in one of three colours of the RGB model is obtained. Obtained satellite images in the red, green and blue spectra are then merged to obtain a multispectral image of the satellite image in the RGB model.

EFFECT: enabling multispectral reconstruction of night images of urban agglomerations based on panchromatic satellite images.

1 cl, 9 dwg