A satellite panchromatic and multispectral image fusion method of a multi-scale convolutional neural network

Abstract

The invention provides a satellite panchromatic and multispectral image fusion method of a multi-scale convolutional neural network. The method comprises the following steps: constructing a basic convolutional neural network, constructing a required training set and a required test set, sharpening the multispectral image, performing data enhancement processing on the training set and the test set,constructing a multi-scale convolutional neural network model, and averaging fusion results of the multi-scale convolutional neural network model to finally obtain a fusion result image; the three-layer convolutional neural network is constructed by selecting a construction method of the three-layer convolutional neural network with simple construction; Compared with the prior art, the method hasthe advantages that the implementation probability of the convolutional neural network is improved, the fusion efficiency is improved, when the network input is formed, the multispectral image is sharpened firstly, the spatial detail information in the multispectral image which is easily ignored in a traditional fusion method can be fully utilized, and the fusion effect of the spatial detail information is improved.

Description

technical field [0001] The invention relates to the field of image fusion, in particular to a multi-scale convolutional neural network satellite panchromatic and multi-spectral image fusion method. Background technique [0002] In recent years, satellites equipped with panchromatic and multispectral image sensors have been launched one after another, and the fusion of panchromatic and multispectral images has become an important and active research direction in the field of satellite remote sensing. This is because in the field of remote sensing, multispectral images with high spatial resolution and high spectral resolution are often required. Image fusion technology is to integrate the characteristics and advantages of panchromatic and multispectral images, and use high spatial resolution panchromatic Image to improve the spatial resolution of multispectral images, while keeping the spectral characteristics of multispectral images as much as possible, so as to overcome the ...

AI Technical Summary

Problems solved by technology

[0003] Regarding the fusion method of panchromatic and multi-spectral images, sub-substitution method is one of the simple and effective classical methods. Common sub-substitution methods include HIS transformation method, PCA transformation method, HCS transformation method and GS transformation method, etc. This type of method is simple to calculate, has real-time and high efficiency, and can improve the spatial resolution of multispectral images well, but when the panchromatic image has little correlation with the components or components to be replaced, it is easy to generate spectra in the fusion image Distortion phenomenon; Another classic fusion method is multi-resolution analysis method, such as Laplacian pyramid, wavelet transform, Contourlet transform, etc. Multi-resolution analysis method is due to the detailed information of panchromatic image and ideal high spatial resolution multispectral The detail components of the image in each channel are not exactly the same, so it is often necessary to further adjust it when blending. Compared with the component replacement method, this method can better maintain the spectral information of the image, but it is difficult to extract and integrate the detailed information. In the process, it is easy to produce blur or distortion, resulting in the loss of spatial information; there is another kind of classical fusion method based on sparse reconstruction fusion method, which randomly samples from other known high spatial resolution multispectral images to construct A multispectral image dictionary with high spatial resolution, however, in practical applications, the multispectral image itself with high spatial resolution is the desired fusion result image, which is not easy to obtain, and there is a problem of huge amount of calculation

Images