Sensitive targets in remote sensing image must be concealed for security. Traditional methods depend on labor-intensive manual processing, which proves both costly and inefficient. This limitation has driven the development of deep learning-based image inpainting techniques as an automated solution. These techniques are specifically designed to replace masked regions while preserving overall image fidelity and structural consistency. Existing inpainting approaches frequently exhibit undesirable artifacts, including ghosting effects and blurry textures, compromising the quality and reliability of the results. To overcome the shortcoming, this paper proposes a novel approach based on diffusion models. Our proposed model effectively replaces foreground targets by utilizing semantic information extracted from the background. The experiments demonstrate that the proposed approach achieves superior performance.