In infrared images, the prominent display of target objects combined with the rich texture details in visible light images can effectively enhance the information entropy of the fused image, thereby providing important support for downstream visual tasks such as nighttime intelligent driving. However, existing mainstream fusion algorithms lack targeted research on the contradiction between the low information entropy and high pixel intensity of visible light images under adverse nighttime road lighting conditions. Therefore, fusion algorithms that perform well under normal conditions often can only generate fused images that are similar to visible light images and have low information entropy under strong light interference. To address this issue, this paper proposes an image fusion network capable of resisting interference from adverse lighting environments, combining information entropy and principles of information theory to enhance the robustness and information retention capability of image fusion. Specifically, we first design an image fusion network with high robustness and excellent performance under normal lighting conditions. Based on this fusion network, we introduce an ambient light perception module to analyze the feature weights of low-information-entropy visible light images under extreme lighting conditions. Additionally, we design an infrared edge feature hierarchical guided fusion module to fully extract effective feature information from infrared images. Experimental results show that this fusion network can fully utilize the feature information of visible and infrared images under adverse nighttime lighting conditions, significantly improving the quality of fused images in such scenarios. Compared with other mainstream algorithms, the fusion results generated by this method contain richer and more effective information.