Satellite video super-resolution reconstruction method and system based on cyclic neural network

Abstract

The invention provides a satellite video super-resolution reconstruction method and system based on a cyclic neural network, which includes acquiring satellite video images and extracting them frame by frame to form an image sequence in the same area; the extracted image sequence is based on the shape and texture of ground objects Features are used for multi-scale object segmentation as geometric constraints to ensure the integrity of ground objects, and the multi-scale segmented image sequence is obtained, which is recorded as a high-resolution image sequence; the simulated satellite video image is degraded during transmission. The scale-divided image sequence is down-sampled, compressed, and noise-added to obtain a degraded low-resolution image sequence; based on the cyclic neural network, the low-resolution image sequence is used as input and the current frame image is used as output to construct deep learning Network; optimize the network, and use the trained cyclic neural network to output super-resolution reconstructed video images. The invention ensures the integrity of ground objects and improves the reconstruction accuracy of satellite video images.

Description

technical field [0001] The invention belongs to the technical field of remote sensing satellite video processing, and relates to a satellite video image super-resolution reconstruction method and system based on a cyclic neural network constrained by geometric objects. Background technique [0002] In the field of remote sensing, video satellites are a new type of earth observation satellites, which are mainly realized by low-orbit video imaging satellites or agile imaging satellites. The biggest difference between the video satellite and the traditional optical remote sensing satellite is that it has a higher time resolution, can continuously observe a certain area, and obtain more information about the movement of the target in the form of video recording, which is especially suitable for moving targets. The high-resolution observation of the target can obtain the moving speed and direction of the target, and this important information is difficult to obtain by traditional...

AI Technical Summary

Problems solved by technology

The current super-resolution reconstruction is based on a single pixel, and does not consider the shape, texture and other characteristics of the ground object. Therefore, based on the current super-resolution technology to perform super-resolution reconstruction of satellite video, the reconstructed satellite video image is prone to The deformation of the moving target leads to poor super-resolution reconstruction

Images