A deep learning network training method for video satellite super-resolution reconstruction

Abstract

The invention discloses a deep learning network training method for super-resolution reconstruction of video satellites. First, a training sample set composed of high-resolution static satellite images is constructed, and then a CNN network structure for super-resolution reconstruction is constructed and network training is set. parameters, and finally establish the loss function for deep CNN training. The method of the invention takes into account the influence degree of the target edge and the pixel gray value on the reconstruction error measurement, thereby improving the training effect of the deep CNN network, and finally promoting the improvement of the performance of the image super-resolution method based on deep learning.

Description

technical field [0001] The invention belongs to the technical field of machine learning and relates to a deep learning network training method, in particular to a deep learning network training method for video satellite super-resolution reconstruction. [0002] technical background [0003] The video satellites emerging in recent years provide an effective means for real-time observation of large dynamic targets by collecting continuous dynamic video, thus greatly making up for the lack of dynamic observation capabilities of traditional remote sensing satellites. The improvement of temporal resolution of video satellites is at the expense of spatial resolution. Generally speaking, the spatial resolution of video satellites is lower than that of remote sensing satellites that perform static or sequence image operations in the same period. For example, the ground resolution of the static image of the optical star on my country's "Jilin-1" reaches 0.72 meters, while the ground ...

AI Technical Summary

Problems solved by technology

Although the definition of this loss function is simple to calculate, it also has obvious limitations in the super-resolution application of video satellites.

First of all, it does not give special consideration to the edges of objects in satellite images, but in fact, enhancing the definition of edge contours of different types of objects has greater value for the interpretation of satellite images

Secondly, according to the brightness masking effect, the perceivable distortion (or reconstruction error) of pixels with different gray levels is different, and the higher the gray value, the greater the distortion is allowed, and vice versa; but the conventional MSE The error measurement function calculates the absolute error, and does not consider the relative influence of the grayscale of the pixel itself

Images