Dual Discriminator Generative Adversarial Imputation Network for Electromagnetic Data Completion

In a battlefield scenario, the surveillance of radar signals within a specified area is crucial for intelligence gathering and tactical decision-making. By analyzing these signals, it is possible to discern the operational mode and objectives of the radar systems. However, due to environmental interference and equipment malfunctions, there is no guarantee that every monitoring platform will be able to obtain real-time and accurate information. Consequently, it is essential to tackle the challenge of using partially observed real data to fill in the gaps. This paper explores the use of the inherent distribution characteristics of monitoring data among fixed monitoring platforms within a fixed area. We propose the application of the Generative Adversarial Imputation Nets (GAIN) to address this issue. To enhance the accuracy of the completion results, we have incorporated the design principles of the Dual Discriminator Generative Adversarial Nets to develop the Dual Discriminator Generative Adversarial Imputation Nets. Compared to GAIN, the Dual Discriminator Generative Adversarial Nets effectively reduces the distribution discrepancy between the generated data and the real data. Experimental results demonstrate the effectiveness of this method in completing missing data for missing rates of up to 60%.