FFMA-Net: Dehazing network via feature fusion guided by mixed attention

Visual perception enhancement in foggy environments is one of the research hotspots in the fields of optics and computer vision today. It obtains clear images by analyzing the feature information of target objects or scenes in foggy images. However, the current popular dehazing algorithms require a large amount of foggy datasets collected by RGB cameras for model training, which is time-consuming and underutilized for feature information, reducing the practicality of the model. In response to this issue, this paper proposes an image dehazing network (FFMA Net) model based on a selected image dataset and mixed attention guided feature fusion. This model combines convolutional attention and Self Attention through a mixed attention module MA and a multi-scale feature fusion module MSFF, maximizing the use of feature information in the image to reconstruct the real scene. The experimental results show that the dehazing network model proposed in this paper exhibits superior dehazing performance in both qualitative and quantitative indicators compared to existing methods on both selected and publicly available datasets.