<p>Remote sensing images are a key data source for recording information about target features. However, they are often contaminated by noise, reducing their usability due to the limitations of imaging equipment, environmental factors, and transmission conditions. In recent years, deep learning-based denoising methods have made significant progress; however, they still have limitations in recovering image details. Therefore, we propose a novel adaptive feature extraction variational interaction denoising method. Firstly, we design an adaptive feature extraction module to tailor the feature extraction process according to the characteristics of remote sensing images and noise levels, capturing the global characteristics of the images at multiple scales. By introducing a priori knowledge to constrain the solution space, we construct a variational estimation sub-network to estimate noise parameters of pixels at different scales according to the variational Bayesian model, capturing complex noise distributions. Combined with a deep interaction sub-network, the model efficiently transfers gradients in deep structures, enhancing the multi-scale representation of image features with detailed information. We conduct extensive experiments on several standard remote sensing image datasets, and the results indicate that our model demonstrates improvements across all evaluation metrics. It effectively avoids the loss of image detail information and retains more high-frequency components, providing a new perspective for the research and application of remote sensing image denoising.</p>