Remote sensing image super-resolution reconstruction method based on fuzzy kernel classification and attention mechanism

Abstract

The invention discloses a remote sensing image super-resolution reconstruction method based on fuzzy kernel classification and an attention mechanism. Firstly, high-resolution and low-resolution optical remote sensing images corresponding to a certain region are given, and a test sample and a training sample are divided; secondly, performing fuzzy kernel estimation on all low-resolution images inthe data; then, fuzzy kernels of all samples in the training set are used for K-means clustering; classifying the high-resolution image pair and the low-resolution image pair of the test set by usinga clustering model; and then constructing a neural network model based on an attention mechanism, setting absolute value errors of the high-resolution image and the low-resolution image as a loss function, obtaining an optimal model according to a test set reconstruction result, finally reconstructing an input image according to the model, and outputting a final result graph. According to the method, the peak signal-to-noise ratio of the reconstructed image can be improved, the robustness is high, and the definition of image edge details is improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a remote sensing image super-resolution reconstruction method based on fuzzy kernel classification and attention mechanism, capable of obtaining a high-resolution image in an optical remote sensing scene. Background technique [0002] The image super-resolution reconstruction method aims to obtain a high-resolution reconstruction result map of the input low-resolution image by processing a single or a series of input images through hardware or software. At present, methods for reconstructing input low-resolution images through software methods can be roughly divided into three categories. The first category is image reconstruction methods based on interpolation, such as: nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, etc.; The second category is based on image reconstruction methods, such as: iterative backprojection, maximum a posteriori ...

AI Technical Summary

Problems solved by technology

However, this method is mainly used in images of natural scenes, and there are still some deficiencies in the super-resolution reconstruction of remote sensing images: First, this method is used in super-resolution reconstruction problems, although the channel attention force mechanism, but does not take into account the spatial position features of the image, and the spatial position features of different sizes can play a key role in estimating the super-resolution reconstruction results of the image. Secondly, when using the channel attention mechanism, the method mainly considers Images in natural scenes, and in remote sensing scenes, due to different sensor parameters and system imaging angles, the degradation models of remote sensing images are also different, and this method does not take into account the different degradation processes that may exist between different input images

Therefore, the above two points limit the reconstruction results of remote sensing images to a certain extent, resulting in a low peak signal-to-noise ratio of the reconstructed remote sensing images, and its reconstruction quality needs to be improved

Images