Unsupervised learning remote sensing image space spectrum fusion method and system

Abstract

The invention provides an unsupervised learning remote sensing image space spectrum fusion method and system, and the method and the system realize remote sensing image fusion through deep learning, and are characterized in that: based on a single group of panchromatic multispectral image pairs, an unsupervised network training mode is adopted to realize fusion of the panchromatic multispectral image pairs; the implementation process comprises the following steps: respectively carrying out downsampling on an originally observed panchromatic image and a multispectral image in a single panchromatic multispectral image pair to serve as network training data pairs, and taking the originally observed multispectral image as network label data to quickly train a fusion network; and inputting theoriginally observed panchromatic image and the multispectral image into the trained fusion network to obtain a fusion image. According to the invention, a downsampling method is used to construct a training data pair, the requirement of a traditional network method training process for a ground truth value is avoided, and unsupervised learning is realized; in addition, for a single-group panchromatic multispectral image pair training network, a large amount of training data is not needed, and network training can be completed quickly.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and relates to a remote sensing image fusion method, in particular to a non-supervised learning space-spectrum fusion scheme of remote sensing images. Background technique [0002] Due to the limitation of satellite sensor hardware, the incident energy that the sensor can receive is limited, resulting in the incompatibility of the spatial-spectral-temporal resolution of a single remote sensing image. At present, many satellites can provide panchromatic and multispectral images of the same scene on the ground at the same time; the panchromatic image has high spatial resolution, but only one band, and the spectral resolution is low; the multispectral image has multiple bands, but the spatial resolution is relatively low. Low, often can not meet the application requirements. Therefore, it is of great research and application value to make full use of remote sensing image fus...

AI Technical Summary

Problems solved by technology

[0004] However, most of the existing deep learning-based methods use supervised learning methods to train and test on simulated data, and use the simulated ground truth (that is, the ideal fusion result) as network label data to supervise the network training process; while in the real situation In , the ground truth does not exist

In addition, the existing deep learning-based methods rely on a large number of training samples. To a certain extent, the sample size determines the accuracy of the fusion network, which makes the training of the network take a lot of time, so it is necessary to study new fusion schemes

Images