Transfer learning-based hyperspectral image super-resolution method

Abstract

The invention discloses a transfer learning-based hyperspectral image super-resolution method, which mainly solves the problem that a deep neural network cannot be trained due to excessively less training data of an original hyperspectral image, improves spatial identification of the hyperspectral image, and solves the problem of spatial resolution limitation of the hyperspectral image. The method mainly comprises the steps of (1) training a convolutional neural network in a natural image database, and learning a mapping relationship between a low-resolution image to a high-resolution image; (2) on a tested hyperspectral image, generating the corresponding high-resolution image waveband by waveband according to the learnt deep neural network; (3) establishing co-matrix decomposition and performing same ground object restriction on the low-high-resolution images; and (4) reconstructing a super-resolution hyperspectral image. According to the method, spectral information of the super-resolution image is kept to the maximum extent; and the method can be used in the fields of remote sensing ground object observation, military reconnaissance, assistance in criminal investigation and the like.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a hyperspectral image super-resolution method. Background technique [0002] With the development of remote sensing technology and imaging spectrometers, the application of hyperspectral remote sensing images is becoming more and more extensive. Hyperspectral remote sensing images have many spectral bands, which sacrifice spatial resolution for high spectral resolution. difficulty. For example, the spatial resolution of the image is low, and the visual recognition is low, resulting in a serious mixture of ground objects, that is, there are pixels mixed with various types of ground objects. It can be said that the spatial resolution has become the main limiting factor for the application effect of hyperspectral images. Therefore, it is very necessary to improve the resolution of the hyperspectral image and enhance the recognition of the objects in the image while keepi...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned existing methods, and propose a hyperspectral image super-resolution method based on migration learning, which can solve the problem that the original hyperspectral image training data is too small and cannot train a deep neural network, and improve hyperspectral image quality. The spatial discrimination of images can overcome the problem of limited spatial resolution of hyperspectral images

Images