网络安全与数据治理2025，Vol.44Issue(11)：30-37,8.DOI:10.19358/j.issn.2097-1788.2025.11.006
基于改进U-Net的多尺度层级融合去雾网络
Multi-scale hierarchical fusion dehazing network based on enhanced U-Net
季云云 1熊亮2
作者信息
- 1. 马鞍山学院 大数据与人工智能学院,安徽 马鞍山 243100
- 2. 国轩高科动力能源有限公司,安徽 合肥 230041
- 折叠
摘要
Abstract
In computer vision tasks,the presence of haze leads to a degradation in image quality.Existing dehazing methods fre-quently suffer from residual haze and color distortion due to insufficient joint modeling of global and local features and the lack of adaptive processing for multi-scale haze concentrations.The proposed multi-scale hierarchical fusion dehazing network based on enhanced U-Net captures hierarchical features through multi-scale inputs and refines information transmission via hierarchical fea-ture fusion.Adjacent layers embed feature enhancement blocks to adaptively focus on key regions,while cross-layer cross-fusion achieves multi-scale complementarity,facilitating full internal information flow.Experiments demonstrate that the method outper-forms comparative approaches on both synthetic and real-world datasets,particularly in detail recovery,color fidelity,and gener-alization to real scenes.关键词
图像去雾/U-Net/多尺度/特征融合Key words
image dehazing/U-Net/multi-scale/feature fusion分类
计算机与自动化引用本文复制引用
季云云,熊亮..基于改进U-Net的多尺度层级融合去雾网络[J].网络安全与数据治理,2025,44(11):30-37,8.
