A remote sensing image fusion method based on a dual-channel neural network

Abstract

The invention discloses a remote sensing image fusion method based on a dual-channel neural network, and the method comprises the steps of firstly, extracting the spatial detail information in a high-frequency component after the high-pass filtering of a panchromatic image through the joint learning of a dual-channel network by utilizing the ARSIS thought; then injecting spatial detail informationinto each waveband image of the multispectral image by using a detail injection model to obtain a required high-resolution multispectral image. According to the method, the advantages of deep learning are effectively utilized, the spatial detail information, different from multi-spectral image waveband images, of a panchromatic image is obtained through double-channel network joint training, andthe correlation between the interior of the image and the images is fully utilized, so that the detail reconstruction is more accurate. Meanwhile, the number of injected details is effectively controlled by using the detail injection model, so that the spatial information storage performance of the fusion algorithm is remarkably improved, and the spectral characteristics of the original multispectral image are well kept.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a remote sensing image fusion method based on a dual-channel neural network. Background technique [0002] With the development of remote sensing technology, remote sensing image data obtained by various satellite sensors provides abundant resources for human observation of the earth, and has made important contributions to a deeper understanding of the world. For remote sensing images, their spatial and spectral resolutions contradict each other. Panchromatic images have high spatial resolution, but less spectral information, and cannot display the color of ground features; multispectral images have rich spectral information, but lower spatial resolution. By fusing panchromatic images with multispectral images, multispectral images with both high spatial and high spectral resolutions will be obtained to meet human needs. [0003] At present, remote sensing...

AI Technical Summary

Problems solved by technology

[0005] In order to make full use of the advantages of deep learning, as well as the structural characteristics and correlations existing within and between images, to improve the accuracy of image fusion, the present invention proposes a The remote sensing image fusion method (DCCNN) based on dual-channel neural network adopts the idea of ​​ARSIS, extracts the spatial detail information from the panchromatic image through the deep learning model, and then uses the detail injection model to inject the detail information into each band of the upsampled multispectral image In order to obtain the required high-resolution multispectral image, solve the problems of low quality and low fusion efficiency of remote sensing image fusion in the prior art

Images