VIFNet introduces a pioneering approach to environmental perception challenges, particularly in image dehazing, by exploiting the strength of visible and infrared data fusion. The network’s multi-scale Deep Structure Feature Extraction (DSFE) module and Channel-Pixel Attention Block (CPAB) significantly enhance spatial and marginal information restoration within deep structural features.

• Proposes the VIFNet framework with a novel multi-scale DSFE.
• Combines visible and infrared data using an inconsistency weighted fusion strategy.
• Demonstrates outperformance through extensive experiments on both real and simulated datasets.
• Makes available a new visible-infrared multimodal dataset called AirSim-VID.
• Enhances robustness of image dehazing in dense-haze conditions.

The significance of VIFNet lies in its ability to maintain environmental perception where traditional single modality methods falter, particularly in dense-haze scenarios. By fully leveraging infrared’s rich information and visible data’s comprehensiveness, the network marks a significant advancement in image processing. Such innovation opens possibilities for applications in areas like autonomous driving, where environmental perception is critical.