During the acquisition of remote sensing images, atmospheric conditions such as clouds often introduce hazy effects, which degrade image quality and impede subsequent analysis. To address this issue, image dehazing algorithms serve as a critical preprocessing step for enhancing remote sensing data. This paper proposes a novel remote sensing image dehazing network that integrates knowledge distillation and frequency-domain features, leveraging adaptive two-branch fusion feature learning. The proposed framework employs frequency-domain mapping to achieve global and local feature fusion, followed by an adaptive attention mechanism to combine frequency-domain and spatial-domain features. Additionally, high-quality remote sensing images are utilized for self-supervised pre-training, and the learned features are transferred to the dehazing network via knowledge distillation. Experimental results, including quantitative metrics and visual assessments, demonstrate the effectiveness of the proposed method in restoring hazy remote sensing images.