Wave Height Inversion Algorithm Based on Multimodal Data Fusion and Attention Mechanism

A wave height inversion algorithm that integrates multimodal data and attention mechanism is proposed by combining shore based monitoring videos and CNN, which can effectively utilize monitoring videos to automatically detect wave height and update wave monitoring technology. This algorithm designs a parallel depth separable convolution module with image denoising function, reducing the algorithm's dependence on high-quality data samples; Propose a multimodal data fusion module based on self-attention mechanism to compensate for the problem that single video image information cannot effectively handle the complex changes of ocean waves. Aiming at the problems of uneven quality of existing datasets and lack of introduction of meteorological data, build the “XiaoMai Island Wave Level Annotation Dataset”. The results show it achieves a Top-1 of 91.3% on Maidao dataset, which is 5.1 % higher. This indicates that the algorithm can effectively detect wave levels and improve algorithm performance.