基于联合训练的红外与可见光图像配准融合算法OA
Infrared and Visible Image Registration Fusion Algorithm Based on Joint Training
红外与可见光图像融合通过异源多模态图像的信息交互来实现全场景的信息互补式增强.针对现有融合方法无法有效处理未对齐图像对问题,文中提出了一种基于联合训练的红外和可见光图像配准融合算法,将图像融合和图像配准视为同一个任务.主干网络采用损失回传结构,使融合图像扮演反馈者的角色融合损失传至配准网络,并在配准阶段引入由粗到细的配准策略来逐步提高配准网络的精度.在融合网络中加入注意力模块AM(Attention Modul),以此实现自适应的特征融合.多基准数据集的定量和定性实验结果表明,所提算法的性能和图像融合质量优于未采用联合训练的同类配准融合算法.
The fusion of infrared and visible light images achieves complementary information enhancement in the entire scene through the information interaction of heterogeneous multimodal images.In view of the problem that the existing fusion methods cannot effectively handle unaligned image pairs,this study proposes an infrared and visible light image registration and fusion algorithm based on joint training,treating image fusion and image registra-tion as the same task.The backbone network adopts a loss-back structure,enabling the fused image to play the role of a feedback unit.The fusion loss is transmitted to the registration network,and a coarse-to-fine registration strate-gy is introduced in the registration stage to gradually improve the accuracy of the registration network.The AM(At-tention Modul)is added to the fusion network to achieve adaptive feature fusion.The quantitative and qualitative ex-perimental results on multiple benchmark datasets show that the performance and image fusion quality of the pro-posed algorithm are superior to those of similar registration and fusion algorithms without joint training.
邢浩杰;杨旭
浙江理工大学 信息科学与工程学院,浙江 杭州 312000浙江理工大学 信息科学与工程学院,浙江 杭州 312000
信息技术与安全科学
红外和可见光图像图像配准图像融合联合训练注意力机制编解码器结构深度学习
infrared and visible imagesimage registrationimage fusionjoint trainingattention mecha-nismcodec structuredeep learning
《电子科技》 2026 (10)
71-77,7
水声技术全国重点实验室稳定支持计划(JCKYS2024604SSJS003) Stable Support Plan for National Key Laboratory of Uunderwater Acoustic Technology(JCKYS2024604SSJS003)
评论