Single-stage denoising fusion collaborative method for fixed structure noise

CN122243793APending Publication Date: 2026-06-19HUAZHONG AGRI UNIV

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: HUAZHONG AGRI UNIV
Filing Date: 2026-03-06
Publication Date: 2026-06-19

AI Technical Summary

Technical Problem

In existing infrared-visible image fusion methods for handling fixed-structure noise, denoising and fusion are difficult to coordinate, resulting in unclear or obscured target information and a decline in the quality of the fused image.

Method used

A single-stage denoising fusion collaborative approach is adopted, including a structural feature extractor, a noise-aware cross-modal feature dictionary complementarity module, and an adaptive collaborative fusion-de-devouring module. Image fusion is performed through a spiking neural network and a multimodal feature encoder. Noise suppression and structural enhancement are achieved by introducing a conditional PatchGAN discriminator and self-supervised constraints. The noise distribution is perceived by a distillation architecture. Image enhancement features are obtained through multi-scale parallel encoding and decoding. Modal complementary feature extraction and spatial alignment are achieved through a fusion mechanism guided by complementary dictionary representation blocks and noise masks.

Benefits of technology

It effectively removes fixed structure noise, improves the accuracy and quality of infrared-visible image fusion, reduces noise content, and improves the PSNR and fusion performance of the image.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122243793A_ABST

Patent Text Reader

Abstract

This invention discloses a single-stage denoising and fusion collaborative method for fixed-structure noise, comprising: constructing a structure refiner based on a spiking neural network to obtain image noise distribution; performing soft noise suppression and enhancing structural edges through generative adversarial training; obtaining common / individual features of the enhanced image through multi-scale parallel encoding and decoding; providing noise range guidance for the denoising task by distillation-guided noise perception enhancement module, reducing the impact of high-frequency structural noise; constructing a complementary dictionary representation block, extracting complementary feature pools between infrared and visible light images through cross-modal interaction to enhance denoising and fusion effects; and constructing an adaptive collaborative fusion denoising module, achieving high-quality fusion of VIS and IR images under noise interference conditions through a noise mask-guided fusion mechanism and aligner. This invention reduces noise content by 89.98%, and compared with state-of-the-art methods, improves infrared thermal imaging perception accuracy by 33.6% and infrared and visible light image fusion accuracy by 10.67%.

Need to check novelty before this filing date? Find Prior Art