CCHD: Chain Connection and Hybrid Dense Attention for Remote Sensing Dehazing

Sen Lin, Mengzhao Liu, Shiben Liu

Abstract: In the process of dehazing of remote sensing images, the problems of haze residue and color overcompensation often occur. To address these problems, we propose a novel remote sensing dehazing network based on chain connection and hybrid dense (CCHD) attention. First, we propose a hybrid attention block (HAB) used to enhance the network’s ability to extract intrinsic structural and textural features of remote sensing images. We design the chain connection mechanism (CCM) to achieve effective feature information transmission, which has the advantage of expanding the network width while reducing the parameters. Then, we develop the feature fusion block (FFB) for deep fusion of multilevel and multiscale remote sensing image feature maps. Finally, we use a gradient-guided block (GGB) to recover edge details and reduce geometric deformation. The experimental results show that our dehazing algorithm performs state-of-the-art methods in both subjective visual evaluation and objective evaluation indexes.

External IDs:doi:10.1109/lgrs.2025.3569580