A single-stage hyperspectral image defogging and reconstruction method based on polarization spectrum combined with prior

The single-stage polarization spectral reconstruction neural network (PST) solves the technical problem of heavy scattering spectrum in existing technologies, achieving efficient dehazing and reconstruction of hyperspectral images. It also solves the problems of insufficient reconstruction accuracy and noise accumulation in existing technologies, and has the advantages of physical interpretability and low cost.

CN122312907APending Publication Date: 2026-06-30NANJING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
NANJING UNIV
Filing Date
2026-04-01
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing hyperspectral imaging techniques suffer from insufficient reconstruction accuracy in hazy environments, complex processing procedures, and noise accumulation issues. Traditional methods have failed to effectively utilize polarization characteristics for physical decoupling.

Method used

A single-stage polarization spectral reconstruction neural network (PST) is employed, combined with a polarization spectral joint imaging model and a U-Net architecture. Through a polarization spectral aggregation attention module and a gated deep convolutional network, end-to-end hyperspectral image dehazing and reconstruction are achieved.

Benefits of technology

It improves computational efficiency, enhances the spectral fidelity and spatial texture details of reconstruction results, provides physical interpretability, reduces hardware costs, and is suitable for fields such as autonomous driving, remote sensing monitoring, and industrial inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122312907A_ABST
    Figure CN122312907A_ABST
Patent Text Reader

Abstract

This invention provides a single-stage hyperspectral image dehazing and reconstruction method based on joint polarization spectral priors, comprising the following steps: Step 1, haze imaging modeling; Step 2, acquiring joint polarization spectral compressed observation data; Step 3, constructing a single-stage polarization spectral reconstruction neural network; Step 4, network training and optimization. The integrated reconstruction architecture proposed in this invention effectively overcomes the limitations of traditional "two-stage" methods. By directly establishing an end-to-end mapping from compressed observations to clear hyperspectral images, it avoids the cumbersome computational steps involved in dehazing followed by reconstruction (or vice versa), significantly improving inference efficiency.
Need to check novelty before this filing date? Find Prior Art