Hyperspectral image super-resolution reconstruction method and device based on depth residual network

Abstract

The invention discloses a super-resolution reconstruction method of a hyperspectral image based on a depth residual network. The method of the invention utilizes a pre-trained depth residual network to carry out super-resolution reconstruction of a hyperspectral image; The depth residual network comprises 2M identical residual blocks, each residual block comprises at least two convolution layers,the super parameters of each residual block are identical, and the weight value is shared, and M is an integer greater than 1; In the forward propagation process of the depth residual network, every 2j residual blocks are grouped as a group respectively, and a jump connection is introduced for each group of residual blocks, j=1, 2,..., M. The invention also discloses a hyperspectral image super-resolution reconstruction device based on a depth residual network. The invention can effectively alleviate the problems of small training samples of hyperspectral images, large amount of single sampledata, difficulty in training, and the like, and to a certain extent, overcomes the limitation of hardware manufacturing technology and imaging environment on the resolution of hyperspectral images.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to View More

AI Technical Summary

Benefits of technology

This patented technology improves upon existing methods for acquiring and analyzing hypergraphic imagery by adding an extra layer called Deep Residue Network (DRN). These networks improve on traditional techniques like denoising autoencodings or histograms that use spectral features alone without relying solely on depth values. By connecting these DRN layers together with a binary linear transformation technique, it becomes possible to efficiently handle both small amounts of data and complex structures within each frame's content. Overall this results in better quality geospatial analysis compared to current approaches.

Problems solved by technology

Technological Problems In this patented method for improving the performance of hypersonic (ultrasonics) radar systems include limitations such as limited space availability due to their costliness, difficulty with acquiring data at different weather environments, difficulties in obtaining accurate maps over vast areas containing materials that may be hidden within them, challenges associated with achieving precise analysis across multiple scenes, potential issues related to environmental effects like temperature fluctuations during flight operations, and complexities involved when analyzing hyperspace radars through various means.

Images