基于深度学习的红外与可见光图像融合方法  

Infrared and Visible Image Fusion Method Based on Deep Learning

作　　者：谢春宇 徐建 李新德[1] 吴蔚[2] XIE Chunyu;XU Jian;LI Xinde;WU Wei(School of Automation,Southeast University,Nanjing 210096,China;Science and Technology on Information Systems Engineering Laboratory,Nanjing 210007,China)

机构地区：[1]东南大学自动化学院,南京210096 [2]信息系统工程重点实验室,南京210007

出　　处：《指挥信息系统与技术》2020年第2期15-20,38,共7页Command Information System and Technology

基　　金：国家自然科学基金(61701454);江苏省自然科学基金(BK20160147和BK20160148);信息系统工程重点实验室开放基金(05201905)资助项目。

摘　　要：红外与可见光图像融合在机器人感知领域中扮演着重要角色,而融合的关键是通过适当方法从源图像中提取有用信息,因此提出了基于深度学习的红外与可见光图像融合方法。首先,将源红外图像分割为前景部分和背景部分,从而构建了基于神经风格迁移算法的红外与可见光图像融合网络;然后,利用前景损失和背景损失分别控制前景和背景2个部分的融合;最后,将2个部分组合在一起重建融合图像。试验结果表明,与其他方法相比,该方法既保留了目标的显著信息,又保留了背景的丰富细节纹理信息。Infrared and visible image fusion is playing an important role in the robot perception,and the key of the fusion is to extract useful information from source image by appropriate methods.A deep learning method for infrared and visible image fusion is proposed.Firstly,the source infrared im‑age is segmented into the foreground part and the background part,and an infrared and visible image fusion network on the basis of neural style transfer algorithm is built.Then,the foreground loss and the background loss are used to control the fusion of the two parts,respectively.Finally,the fused image is reconstructed by combining the two parts.The experimental results show that compared with other existing methods,the method retains both saliency information of target and detailed texture information of background.

关 键 词：红外图像 可见光图像 图像融合 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]