Multi-weather degradation image restoration method based on state space model

By combining a state-space model with convolution and adaptive channel attention modules, this image restoration method addresses the issues of global receptive field and computational efficiency in image restoration under various weather conditions. It achieves efficient and lightweight image restoration results, improving the robustness and adaptability of the model.

CN119295348AActive Publication Date: 2025-01-10SHENYANG INST OF COMPUTING TECH CO LTD THE CHINESE ACAD OF SCI
View PDF 5 Cites 4 Cited by

Patent Information

Application Number
CN202411341849.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-25
Publication Date
2025-01-10
Estimated Expiration
2044-09-25

AI Technical Summary

Technical Problem

Existing image restoration techniques struggle to balance global receptive field and computational efficiency when dealing with various weather-degraded images. Traditional methods are computationally expensive or have limited receptive fields in resource-constrained environments, and require different model parameters or prediction training weights under different weather conditions, resulting in excessive storage and computational burdens.

Method used

A multi-weather degraded image restoration method based on a state-space model is adopted. Combining the local feature extraction of convolution and the long-range dependency capability of the state-space model, the method achieves lightweight and efficient end-to-end image restoration through shallow and deep feature extraction stages, using residual visual state-space groups and adaptive channel attention modules.

Benefits of technology

It achieves efficient image quality restoration under various weather conditions, maintains high computational efficiency, improves the robustness and generalization ability of the model in complex scenes, reduces computational cost and number of parameters, captures multi-scale features, and enhances the accuracy and clarity of image restoration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119295348A_ABST
    Figure CN119295348A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-weather degraded image restoration method based on a state space model. The method comprises the following steps: acquiring multi-scene clear images, synthesizing images under rain, snow and fog weather conditions by using an image synthesis technology, and constructing a standard multi-weather data set; preprocessing the data set and then dividing the data set; constructing a multi-weather degradation image restoration model based on the state space model, and performing training optimization on the model; the method comprises the following steps: acquiring an image which is degraded due to weather influence, preprocessing the image, inputting the processed image into a pre-trained multi-weather degraded image restoration model for restoration, and outputting the image as a processed clear image. According to the method, a self-adaptive channel attention mechanism is introduced based on the state space model, the capturing capability of the network on image local features is enhanced, images of various scales can be effectively processed through overlapped block embedding and multi-scale processing strategies, and the image recovery performance is remarkably improved. The method provided by the invention can process image restoration tasks including but not limited to various weather degradation caused by rain, snow and fog, and has a wide application prospect.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Infrared image super-resolution reconstruction method combining two-way convolution and self-attention

    CN117274047A

  • High-resolution reconstruction method for remote sensing image under complex degradation model

    CN117474764A

  • Weather detection method based on multi-modal data fusion framework

    CN118351380A

  • General image restoration method for severe weather based on two-stage distillation learning

    CN118521511A

  • Lightweight image super-resolution reconstruction method and system, storage medium and product

    CN118552408A

Cited By

  • Image restoration method and system coping with various severe weather environment factors

    CN120355607A

  • Multi-modal sentiment classification method and system based on cross-gating state space fusion

    CN121211223A

  • Multi-modal sentiment classification method and system based on cross-gating state space fusion

    CN121211223B

  • Road disease tracking method and system based on machine vision

    CN122530206A