Image dehazing method based on visual state space model and depth information assistance

Image dehazing, as a low-level visual task, has always been a challenging research topic in computer vision. This paper proposes a dehazing algorithm, UDMDN, based on the visual state space model and augmented with depth information. First, a mamba dehazing network based on U-net is proposed, which has efficient modelling capabilities while reducing computational consumption; second, a gating mechanism and a deep separable convolutional feedforward network are introduced to achieve better feature extraction capabilities; finally, the dehazing network and the depth estimation network are collaborated to use differential information to better guide the dehazing network to focus on the foggy area, and the dehazing effect is further improved by uncertainty loss optimization. Experimental results show that the model proposed in this paper has better dehazing performance than existing models.