Infrared and visible light image fusion method and system and medium

Abstract

The invention discloses an infrared and visible light image fusion method and system and a medium, and the method comprises the steps: respectively inputting an infrared image Ia and a visible light image Ib into a depth feature extraction network D to obtain an infrared up-sampling feature map Da and a visible light up-sampling feature map Db, and splicing the infrared up-sampling feature map Da and the visible light up-sampling feature map Db to obtain a fusion image; respectively inputting the infrared image Ia and the visible light image Ib into a depth feature extraction network D to extract image features of various scales, generating a weight through a mask up-sampling network Umask, and performing weighted summation on a reconstructed image generated through a feature up-sampling network Uoriin to obtain fusion layers of various scales; and inputting the fused image into the feature up-sampling network Uorin, and fusing the fused image with the fusion layers of various scales to obtain a final fused image. The image generated by the method has the advantages of being good in visualization effect, obvious in texture information, high in feature extraction capability, high in structure retention degree, free of obvious artifacts, high in universality and the like.

Description

technical field [0001] The invention relates to infrared and visible light image fusion technology, in particular to an infrared and visible light image fusion method, system and medium. Background technique [0002] Fusion of infrared and visible light images aims to integrate infrared images and visible light images into a more informative and more visualized image, and the generated image can provide effective help for subsequent processing and decision-making. When performing visible light and infrared fusion tasks, we usually pay attention to the difference between the target and the background of infrared images, the texture information of visible light, and whether the fusion image conforms to human visual perception. It has unique advantages in target tracking, target detection, scene analysis and other fields. Infrared and visible light image fusion methods generally include multi-scale transformation, sparse representation, traditional methods such as subspace-bas...

AI Technical Summary

Problems solved by technology

U2fusion can complete infrared and visible light image fusion tasks well at the same time. The network is mainly designed for the loss function, but its network training needs to use multiple sets of task data to jointly optimize the network, and it needs to use pre-trained weights to participate in the training process. For input The image is processed by simple concat (connection), so the network feature extraction process is relatively simple, and the training cost is relatively high

Images