An enhanced multi-linear mixing hyperspectral unmixing method, system and device
By constructing a multi-branch variational autoencoder unmixing network and combining spectral variability and nonlinear mixing models, the problem of limited accuracy in hyperspectral unmixing under complex scenarios was solved, and high-precision endmember extraction and abundance inversion were achieved.
CN122223458APending Publication Date: 2026-06-16SHAOXING UNIVERSITY
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAOXING UNIVERSITY
- Filing Date
- 2026-05-19
- Publication Date
- 2026-06-16
AI Technical Summary
Technical Problem
Existing hyperspectral unmixing methods struggle to effectively handle spectral variability and nonlinear mixing phenomena in complex scenarios, resulting in limited unmixing accuracy.
Method used
A multi-branch variational autoencoder demixing network is adopted, which combines spectral variability and nonlinear hybrid models. Through joint loss function optimization training, the decoupled learning of spectral variability and nonlinear scattering parameters is achieved.
Benefits of technology
It improves the accuracy and robustness of hyperspectral unmixing in complex environments under unsupervised conditions, and achieves high-precision endmember extraction and abundance inversion.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122223458A_ABST
Abstract
The application discloses an enhanced multi-linear mixed hyperspectral unmixing method, system and device, which comprises the following steps: acquiring hyperspectral image data of a region to be processed, and preprocessing the hyperspectral image data; pre-extracting endmember and abundance information from the preprocessed hyperspectral image data; constructing a multi-branch variational autoencoder unmixing network; based on an enhanced multi-linear mixed model considering spectral variability, performing reconstruction calculation on a pixel at the end of the multi-branch variational autoencoder unmixing network; constructing a joint loss function, and iteratively optimizing and training the multi-branch variational autoencoder unmixing network; and outputting an unmixing result, and obtaining a variable endmember spectrum and a corresponding ground object abundance distribution. The application combines the feature representation capability of a deep generative model and the mechanism driving of a real physical model, and can still maintain high endmember extraction and abundance inversion precision under the coupling of spectral variability and nonlinear spectral mixing and complex interference, and thus improves the performance of hyperspectral unmixing in a complex environment.
Need to check novelty before this filing date? Find Prior Art