摘要
本文提出了一种基于坐标注意力机制和梯度残差密集块的融合算法,利用普通卷积提取源图像的浅层特征,使用梯度残差密集块模块提取源图像深层特征和细粒度细节特征;使用坐标注意力模块捕获特征图中空间位置之间的远程依赖关系重,将融合后的特征利用解码器重建融合图像。实验结果说明本文提出的融合算法具有较好的融合性能。
In this paper a fusion algorithm based on gradient residual dense block and attention mechanism is proposed.The shallow features of the source images are extracted by convolution,and the deep features and fine-grained details of the source images are extracted by gradient residual dense block module.The coordinate attention module is used to capture the remote dependence between the spatial positions in the feature maps,and the decoder reconstructs fused features to obtain the fused image.The results show the proposed fusion algorithm has better fusion performance.
作者
耿鹏
吴薇娜
卢琳
Geng Peng;Wu Weina;Lu Lin(Shijiazhuang Tiedao University,Hebei,050000;Shijiazhuang No.24 middle school,Hebei,050000)
出处
《石家庄铁路职业技术学院学报》
2023年第1期65-68,共4页
Journal of Shijiazhuang Institute of Railway Technology
基金
国家自然科学基金项目(61972267)
河北省大中学生科技创新能力培育专项(22E50515D)。
关键词
坐标注意力
图像融合
卷积
损失函数:残差网络
Coordinate attention
image fusion
convolution
loss function
residual network
