红外技术2025，Vol.47Issue(2)：193-200,8.
基于深度学习的偏振图像融合方法
Deep Learning-Based Polarization Image Fusion Method
摘要
Abstract
To improve image quality in complex and dim environments,a network that leverages both the global information and textural details of polarized images through a strategy of multi-scale feature extraction and dual fusion,known as the Scale Feature Extraction and Dual Fusion Strategy Network(SFE-DFS-Nest),is proposed.The proposed network fuses polarized intensity images with polarization degree images.Initially,an encoder is constructed to extract multi-scale features from source images.Then,shallow features are fused using a lightweight Transformer,while deep features are integrated through a residual network.Finally,a decoder is built to reconstruct the fused features.Compared with existing image fusion networks,this network employs distinct fusion strategies for features at different scales.The experimental results show that images from dark and complex environments exhibited improved subjective visual comfort after fusion through this network.Furthermore,the fused images obtained using the proposed method outperformed those obtained using the compared methods in terms of objective evaluation metrics.关键词
SFE-DFS-Nest网络/偏振图像/图像融合Key words
SFE-DFS-Nest network/polarized images/image fusion分类
信息技术与安全科学引用本文复制引用
孙红雨,李军,袁博,周宇超..基于深度学习的偏振图像融合方法[J].红外技术,2025,47(2):193-200,8.基金项目
山东省自然科学基金(ZR2024QF085). （ZR2024QF085）
