现代电子技术2025，Vol.48Issue(23)：1-8,8.DOI:10.16652/j.issn.1004-373x.2025.23.001
基于聚合像素特征提取的Transformer图像去雾算法
Transformer-based image dehazing algorithm with aggregated pixel feature extraction
摘要
Abstract
To address issues in existing image dehazing methods such as incomplete recovery of image detail texture features,neglect of important scene edge information,and poor generalization ability,a Transformer-based image dehazing algorithm with aggregated pixel feature extraction is proposed.Firstly,a dual-path aggregated pixel feature extraction module is introduced,focusing on both fine-grained and coarse-grained features to enhance the extraction of image information in hazy environments.Secondly,on the basis of the characteristics of feature granularity and attention mechanisms,the dual-path module is deeply optimized to aggregate pixel feature extraction,enhancing the ability to restore image details and the fusion of channel features.Finally,to further integrate multi-scale information and enable the network to more thoroughly understand and represent the input data,an adaptive feature fusion module is proposed.The redundant residual connections are removed and the generalization ability of the dehazing model is improved effectively,and the fine texture details are reconstructed.To verify the effectiveness of the proposed method,qualitative and quantitative comparisons were conducted against eight existing dehazing algorithms on three public datasets named RESIDE-IN,RESIDE-OUT,and RESIDE-6k.The experimental results demonstrate that the proposed model excels in dehazing performance,detail texture recovery,and generalization ability,and can be applied to the field of image defogging.关键词
注意力机制/双路径特征提取/聚合像素特征/Transformer/多尺度融合/图像去雾Key words
attention mechanism/dual-path feature extraction/aggregated pixel feature/Transformer/multi-scale fusion/image dehazing分类
电子信息工程引用本文复制引用
尹学辉,陈秋雨..基于聚合像素特征提取的Transformer图像去雾算法[J].现代电子技术,2025,48(23):1-8,8.基金项目
国家自然科学基金项目(61701060) （61701060）
