MRDUnet:Unet-based Multi-level Adaptive Residual Image Desnowing Network

Abstract Snowflakes can obscure important image content and significantly impede information retrieval. Owing to their complex shapes, varying scales, and random spatial distributions, existing desnowing methods often fail to identify snow particles accurately, resulting in residual snowflakes or dark artifacts in restored images. To address these issues, this paper proposes a multi-level adaptive residual desnowing network based on U-Net, termed MRDUnet. First, a snow mask prediction sub-network is constructed within the MRDUnet framework by integrating a Global Snowflake Perception Network (GSPN) into U-Net to improve snow localization and mask prediction. ResNet modules are further incorporated to enhance snow-related feature representation, preserve fine image details, and reduce information loss. Subsequently, Feature Reconstruction Networks (FRNets) are embedded between different layers of the mask prediction sub-network to progressively extract snow-free features at multiple scales. Skip connections are employed to facilitate feature propagation and enrich the reconstructed image representation. Finally, a Multi-Scale Feature Fusion Network (MSFF-Net) integrates the multi-level features and reconstructs the final snow-free image. Extensive experiments on synthetic and real-world datasets demonstrate that MRDUnet achieves state-of-the-art results on Snow100K while remaining competitive on CSD and SRRS, producing visually clear restored images with stable performance across different datasets.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-13
DOI
https://doi.org/10.1088/2631-8695/aec21a
Primary Topic
Image Enhancement Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

MRDUnet:Unet-based Multi-level Adaptive Residual Image Desnowing Network

PeiLin Wang
2 links
Engineering Research Express
Image Enhancement Techniques
article

MRDUnet:Unet-based Multi-level Adaptive Residual Image Desnowing Network

PeiLin Wang
article en

Abstract

Abstract Snowflakes can obscure important image content and significantly impede information retrieval. Owing to their complex shapes, varying scales, and random spatial distributions, existing desnowing methods often fail to identify snow particles accurately, resulting in residual snowflakes or dark artifacts in restored images. To address these issues, this paper proposes a multi-level adaptive residual desnowing network based on U-Net, termed MRDUnet. First, a snow mask prediction sub-network is constructed within the MRDUnet framework by integrating a Global Snowflake Perception Network (GSPN) into U-Net to improve snow localization and mask prediction. ResNet modules are further incorporated to enhance snow-related feature representation, preserve fine image details, and reduce information loss. Subsequently, Feature Reconstruction Networks (FRNets) are embedded between different layers of the mask prediction sub-network to progressively extract snow-free features at multiple scales. Skip connections are employed to facilitate feature propagation and enrich the reconstructed image representation. Finally, a Multi-Scale Feature Fusion Network (MSFF-Net) integrates the multi-level features and reconstructs the final snow-free image. Extensive experiments on synthetic and real-world datasets demonstrate that MRDUnet achieves state-of-the-art results on Snow100K while remaining competitive on CSD and SRRS, producing visually clear restored images with stable performance across different datasets.

Engineering Research Express
Lanzhou Institute of Technology (CN)
Image Enhancement Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon