Research on Task-Driven Dual-Light Image Fusion and Enhancement Method under Low Illumination

In low light situations, a single visible image can not transmit reliable information, even cause the loss of the target information. At this point, the advantages of visible and infrared image fusion will be highlighted. For a given pair of visible and infrared images, they are collectively referred to as dual-light images in this paper. How to make the most of their information and improve the information expression ability of the fused image is crucial. The traditional evaluation methods use statistical indicators, which is not associated with the upstream task. In this paper, the image fusion method driven by the target detection task is studied. Semantic loss is added to guide the dual-light image fusion. Moreover, through the visual enhancement module, the impact of adverse factors ( low light, etc. ) on the image is weakened, and the information expression level of the image is improved. Thus, the final image is more beneficial to target detection.