Deep learning-based oceanic internal wave detection from SWOT satellite imaginary observations

Oceanic Internal Waves (OIWs) are subsurface gravity waves that strongly influence ocean mixing, nutrient transport, and maritime operations. Detecting OIWs from SWOT hydro-logical satellite imagery is challenging due to their subtle stripe-like signatures and frequent interference from wind streaks, ship wakes, and sensor noise. This paper proposes a novel hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) and Transformer encoders with Spatial Reduction Attention (SRA) for accurate and efficient OIW detection. The CNN backbone extracts fine-grained stripe textures, while the Transformer module models long-range dependencies across large wave packets; a multi-scale fusion strategy combines hierarchical features for robust classification. We validate the method on a large SWOT-based dataset containing 31,495 annotated patches across multiple ocean regions. Experimental results demonstrate state-of-the-art performance, achieving 96.2% accuracy, outperforming state-of-the-art baselines. Ablation studies confirm the complementary benefits of CNN-Transformer integration and the efficiency gains of SRA, enabling real-time inference (35 ms per patch) and full-swath processing within seconds. The proposed approach provides a scalable solution for operational ocean monitoring, maritime safety, and long-term internal wave climatology.