Multispectral remote sensing image fusion method and device based on residual learning

Abstract

The invention discloses a multispectral remote sensing image fusion method and a multispectral remote sensing image fusion device based on residual learning. The method comprises the following steps:(1) acquiring a plurality of original multispectral remote sensing images IMS and corresponding original panchromatic band remote sensing images IPAN; (2) calculating to obtain an interpolation imageIMSI of the IMS, a gradient image GPAN of the IPAN and a differential image DPAN; (3) constructing a convolutional neural network fusion model which comprises a feature extraction layer, a nonlinear mapping layer, a residual image reconstruction layer and an output layer which are connected in sequence, taking I=[IMSI,IPAN,GPAN,DPAN] as input for training, and taking a loss function adopted duringtraining as a mean square error function for introducing residual learning; and (4) processing the multispectral remote sensing image I'MSRI to be fused and the corresponding original panchromatic band remote sensing image I'PAN to obtain corresponding data [I'MSI, I'PAN, G'PAN, D'PAN], inputting the data into the trained convolutional neural network fusion model, and outputting an image after fusion. The method is high in fusion speed, and the spectrum and space quality of the fused image is higher.

Description

technical field [0001] The invention relates to image fusion technology, in particular to a multi-spectral remote sensing image fusion method and device based on residual learning. Background technique [0002] With the further development of remote sensing technology, remote sensing images are widely used in fields such as agricultural production, environmental monitoring, and geological survey. However, in actual production applications, due to the structural limitations of remote sensing equipment, remote sensing images with high spatial and spectral resolutions are difficult to obtain directly. To solve this problem, the current remote sensing image acquisition equipment carried by satellites often has two different sensors, which can respectively acquire two kinds of remote sensing images: panchromatic remote sensing images with high spatial resolution and multispectral remote sensing images. Using multi-spectral remote sensing image fusion technology to effectively fu...

AI Technical Summary

Problems solved by technology

Mainly using wavelet transform, high-pass filtering and other methods, although the spectral distortion phenomenon has been improved, but this type of method still has certain limitations in spectral fidelity

3) Hybrid method, mainly by combining the ideas of different methods, taking CS and MRA as an example of guided filter principal component analysis method, although it can combine the advantages of different methods, but the method has obvious limitations, and the performance cannot be improved

Taking the variational method as an example, it has a good improvement on the spectral distortion phenomenon, but the time consumption is greatly increased and it has a strong dependence on the data characteristics.

Images