Multi-scale Generative Adversarial Network-Based Modeling for Radar Data Occlusion Area

We propose a radar data occlusion area modeling method based on multi-scale generative adversarial networks to address the common occlusion phenomenon in radar data in complex environments. The proposed method designs a pyramid shaped multi-scale generative adversarial network, achieving spatial consistency and optimization of statistical features in the obstructed area. We have designed a loss function suitable for simulating radar data occlusion areas, using spatial consistency constraints and high-order moment matching strategies to enhance the rationality and authenticity of the generated areas. To verify the effectiveness of the method, the study used measured data from synthetic aperture radar (SAR) for testing. The results show that the proposed method improves key indicators such as structural similarity and information entropy by 3% to 5% compared to traditional generative adversarial networks. In addition, in the simulation task of synthetic aperture radar interferogram occlusion area, this method improves image details and enhances the imaging integrity and stability of the occlusion area. The comprehensive analysis results indicate that this method not only improves the modeling accuracy of radar data, but also provides innovative solutions for the completion of missing radar data and intelligent imaging.