现代信息科技2025，Vol.9Issue(22)：30-34,39,6.DOI:10.19850/j.cnki.2096-4706.2025.22.006
基于注意力机制的递进式特征提取去雾网络
Progressive Feature Extraction Dehazing Network Based on Attention Mechanism
摘要
Abstract
Aiming at the limitations of existing single image dehazing algorithms in terms of accuracy and detail retention,a progressive feature extraction dehazing network based on attention is constructed.Taking AOD-Net as the benchmark framework,the feature interaction path is reconstructed by pointwise convolution and multi-dimensional collaborative Attention Mechanism,which reduces the scale of model parameters and improves the computational efficiency.The progressive feature extraction network structure is designed,and the multi-scale feature fusion strategy is used to enhance the separation ability of the network to the long-range fog concentration gradient and high-frequency details.The multi-scale structural similarity constraint and the adaptive loss optimization mechanism are further integrated to significantly improve the consistency of texture structure and the balance of color distribution in the restored image.The experimental results show that the proposed network exhibits excellent detail retention ability and visual naturalness in both synthetic and real fog image scenes.关键词
图像去雾/卷积神经网络/注意力机制/多尺度网络Key words
image dehazing/Convolutional Neural Network/Attention Mechanism/multi-scale network分类
信息技术与安全科学引用本文复制引用
程小园,王炳文,李葆光,封蕾,金能智..基于注意力机制的递进式特征提取去雾网络[J].现代信息科技,2025,9(22):30-34,39,6.基金项目
甘肃省科技计划项目(24CXTA002) （24CXTA002）
