DiffWave for Sea Clutter Data Generation

Computer simulated sea clutter generation cuts the high costs and long cycles of field collected sea clutter data. Traditional statistical modeling with real data lacks realism and generalization. GAN and VAE simulation methods struggle to capture global sea clutter features and generate complex sea clutter. To solve these problems, this paper proposes a sea clutter generation method based on DiffWave. We build a one dimensional sea clutter DiffWave model and train it with real data. We analyze the generated clutter’s temporal domain and frequency domain characteristics, as well as its temporal and spatial autocorrelation. The model is compared with existing ones using MMD metrics. Real data validation shows that it can generate more realistic sea clutter data.