Remote sensing image-to-image translation addresses multi-source data integration challenges by establishing cross-domain mapping for content-style disentanglement within a unified feature space, thereby enabling seamless fusion of multi-source data. This technology alleviates the scarcity of training data and high annotation costs, boosting downstream interpretation efficiency. However, existing methods—primarily adapted from computer vision—are constrained by statistical independence assumptions between content and style features. This fundamental limitation overlooks the critical modulatory influence of style attributes on content characteristics, exacerbating conflicts between natural image paradigms and remote sensing physical principles. To resolve these issues, we propose the Style-Conditioned Distribution Translation Network (SCD-TransNet). Our method establishes the physical dependency relationship between content and style attributes by learning style-conditional distribution translation, achieving collaborative optimization of content preservation and style adaptation. Furthermore, we introduce structural similarity constraints in the loss function to enforce semantic consistency. Benefiting from these improvements, our method achieves superior performance in both qualitative and quantitative comparisons with other methods on various datasets. The results of the ablation experiment further validate the effectiveness of the proposed method.