Ocean Wave Information Extraction from Raw Ocean Video Based on Self-Attention ConvLSTM

Precise wave information is essential for assessing fatigue damage to offshore wind turbines. This study explores a visual method based on ConvLSTM combined with CNN and self-attention memory to extract ocean wave details. The network extracts both temporal and spatial information from wave impact videos to estimate significant wave height (Hs) and wave peak frequency (fp). Separate training for Hs and fp demonstrates the network's accuracy under varying imaging conditions. Additionally, the inclusion of a self-attention memory module based on ConvLSTM enhances model performance, improving prediction and regression capabilities.