A New Adversarial Hierarchy for SAR Clutter Generation

Clutter simulation plays an important role in SAR image interpretation. The preceding works usually generate the clutter by the amplitude of measured data. They ignore the phase information. In addition, it is difficult to describe the ground clutter accurately with a certain statistical distribution model, due to the complex and mutable scenes. To address these problems, this paper presents a new adversarial hierarchy for SAR clutter generation. Different from the preceding works, we jointly consider the real and the imaginary components into a unified framework. They are used to generate the complex clutter data. It is therefore no need to tune the model parameters manually. On the other hand, a new measurement, Jensen-Shannon divergence is developed to quantify the similarity between the generated data and the real clutter data. Multiple comparative studies are made to demonstrate the advantages of the proposed framework and the effectiveness of the measurement.