Remote sensing single image super-resolution method based on channel self-attention multi-scale feature learning

Abstract

The invention relates to the technical field of mode recognition, in particular to a remote sensing single-image super-resolution method based on channel self-attention multi-scale feature learning, and aims to solve the problem that an existing image super-resolution method depends on a large number of synthesized external data sets. According to the remote sensing single-image super-resolution method based on channel self-attention multi-scale feature learning, a new image super-resolution network model, namely a channel self-attention multi-scale feature learning network, is provided, and the channel self-attention multi-scale feature learning network does not need any additional synthetic data set; only one low-resolution input image is needed for training; besides, features of different scales of the image are extracted through multiple columns of convolution layers with different convolution kernels, the channel self-attention multi-scale feature learning network uses a well-designed network, the multi-scale features of the remote sensing image are fully learned, and super-resolution of the remote sensing image is achieved; the channel self-attention multi-scale feature learning network improves the super-resolution performance.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a remote sensing single image super-resolution method based on channel self-attention multi-scale feature learning. Background technique [0002] Image super-resolution refers to the reconstruction of a given low-resolution image into a high-resolution image, which is a very ill-posed process. In the past few years, researchers have proposed a large number of super-resolution models to address this ill-posed problem. At present, the main image super-resolution methods are as follows: [0003] (1) Image super-resolution methods based on interpolation: Early image super-resolution methods are mainly based on interpolation, and the nearest neighbor interpolation method, bilinear interpolation method, and bicubic interpolation method are the three most common interpolation-based methods. . The nearest neighbor interpolation method assigns the pixel value closest to the...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a remote sensing single-image super-resolution method based on channel self-attention and multi-scale feature learning to solve the problem of dependence of existing image super-resolution methods on a large number of synthetic external data sets

Images