Application of image completion in remote sensing image desensitization

Remote sensing images are important data for applications such as urban planning and resource surveys. They contain rich geographical and resource distribution information, including sensitive information that should not be made public. Therefore, remote sensing images need to undergo desensitization processing before being publicly available. Traditional desensitization methods for remote sensing images often rely on manual processing, which is not only time-consuming and labor-intensive but also requires a high level of expertise from technical personnel. With the increase in frequency and accuracy of remote sensing image acquisition, the efficiency of traditional desensitization methods no longer meets practical needs. To address this issue, a fully automatic high-fidelity image completion technique based on global contextual interactions is applied to the desensitization task of remote sensing images, and its feasibility has been verified. Experimental results demonstrate that this method achieves high processing efficiency and improves the credibility of desensitized content in medium-scale automatically generated remote sensing images to a certain extent. It provides new ideas and directions for remote sensing image desensitization. In addition, in response to the demand for large-scale remote sensing images, methods such as adding BN layers are proposed to enhance the learning and generalization abilities of the model, thereby obtaining high-quality images.