<p>Remote sensing (RS) image dehazing is a challenging task, as images recorded in hazy weather usually undergo heavy degradation and artifacts. Although advances have been made, current techniques still struggle to balance RS image recovery and computational efficiency.To solve the mentioned challenges, we propose an innovative Dual-Headed Mamba Diffusion Model named DMa-Diff, integrating a novel Mamba variant with diffusion models for efficient RS dehazing. To be specific, the Dual-Headed Mamba Block (DHMB) is designed to integrate spatial features for effective dehazing, which leverages inference efficiency from the State Space Model (SSM). Its unique dual-branch architecture significantly enhances spatial feature extraction, ensuring high-quality image restoration while markedly improving inference efficiency. In addition, Adaptive Frequency Filter (AFF) is employed to address high-frequency details for a realistic dehazed effect. This filter precisely fine-tunes high-frequency information across various degradations, ensuring the authenticity and clarity of the RS dehazed images. We prove our model’s effectiveness through extensive experimental evaluations on both synthetic datasets and real RS imagery.</p>