Noise Reduction in Synthetic Radar Returns via Denoising Autoencoder for Estimating Ocean Wave Parameters

The integration of artificial intelligence (AI) into marine radar systems holds transformative potential for ocean monitoring, particularly in South Korea, where the spatial measurement network for the ocean remains underdeveloped. This study seeks to overcome the limitations of marine radar, especially in mitigating noise and accurately capturing wave field dynamics under both calm and extreme sea conditions. To achieve this, we enhance the radar's 3D Fast Fourier Transform (FFT) using a custom-made denoising autoencoder (DAE) architecture. Although the stereo camera system was initially planned to provide the training ground truth data for the AI model, this study instead tests the AI using synthetic real sea surface datasets to offer greater flexibility in simulating various wave conditions. Wave parameters were extracted from the 3D FFT and analyzed across multiple Beaufort-scale scenarios. The DAE application resulted in substantial noise reduction, with signal-to-noise ratio (SNR) improvements of over 13 dB, thus improving the clarity and accuracy of wave patterns in radar returns. The results highlight the potential of AI-enhanced radar systems to refine wave field analyses, particularly in complex and extreme sea states. Future work will focus on further optimizing the AI architecture for real-world marine radar and stereo camera datasets, advancing its operational readiness for disaster mitigation and oceanographic research.