Sea clutter suppression for radar images based on multi-head self-attention diffusion model

To address the severe interference of sea clutter on radar echo signals, a sea clutter suppression for radar images based on a multi-head self-attention diffusion model is proposed. Based on the diffusion model, a multi-head self-attention diffusion network model (MHA-DNet) is developed by integrating a multi-head self-attention mechanism. which facilitates feature extraction and learning of the characteristics of sea clutter. In this study, diverse targets of ocean clutter time-frequency images are simulated based on the IPIX Radar clutter dataset, resulting in a large-scale and stochastic image dataset. This approach enhances the model generalization ability, making it more robust. On the newly proposed peak signal-to-noise ratio structural similarity index (P-S) evaluation metrics in this paper, MHA-DNet shows a high performance improvement of 4% compared with the traditional convolutional method, and 1.1% compared with the GAN adversarial network method, and even compared with the original diffusion model, MHA-DNet demonstrates a 0.2% advantage, which verifies that the method proposed in this paper has a certain degree of effectiveness in the sea clutter suppression.