A spectral super-resolution method based on variational autoencoder

By using a spectral super-resolution method based on variational autoencoders, RGB images are acquired using ordinary cameras and hyperspectral images are reconstructed. This solves the problems of long imaging time and high equipment specialization in hyperspectral imaging systems, and achieves fast and effective hyperspectral image acquisition.

CN115861071BActive Publication Date: 2026-06-26GUANGDONG UNIV OF TECH +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUANGDONG UNIV OF TECH
Filing Date
2022-12-19
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Hyperspectral imaging systems require acquiring a large number of narrow spectral bands within a set spectral range, which takes a long time and requires specialized equipment, making it difficult to efficiently acquire hyperspectral images.

Method used

A spectral super-resolution method based on variational autoencoders is adopted. RGB images are acquired through ordinary cameras, and hyperspectral images are reconstructed using a spectral super-resolution network and attention mechanism module, including shallow feature extraction, feature mapping and hyperspectral image reconstruction.

Benefits of technology

It reduces the difficulty of acquiring hyperspectral images, shortens imaging time, improves imaging efficiency, and enables rapid reconstruction of hyperspectral images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115861071B_ABST
    Figure CN115861071B_ABST
Patent Text Reader

Abstract

The application discloses a spectral super-resolution method based on a variational autoencoder, and comprises the following steps: inputting a target RGB image into a first convolutional layer of a spectral super-resolution network to obtain a plurality of feature maps containing shallow layer features corresponding to the target RGB image; inputting the feature maps into an attention mechanism module of the spectral super-resolution network to perform feature mapping on the feature maps; inputting the feature maps subjected to the feature mapping into a second convolutional layer of the spectral super-resolution network to reconstruct a target hyperspectral image corresponding to the target RGB image; and training the spectral super-resolution network according to a first hyperspectral image and a first RGB image extracted by the variational autoencoder, wherein the first hyperspectral image is reconstructed by the untrained spectral super-resolution network on the basis of a training RGB image. The target RGB image is acquired by a common camera, so that the image acquisition difficulty is low, and the hyperspectral image is reconstructed by the spectral super-resolution network, so that the imaging time is short, and the method can be widely applied to the field of spectral super-resolution.
Need to check novelty before this filing date? Find Prior Art