Tri-GACNet: A Storm Surge Point Cloud Segmentation Algorithm Based On Three-Channel Graph Attention Convolution

Storm surge is one of the most important marine hazards in China, and in the previous numerical calculation of storm surge volume, the model is less effective in calculating the near-shore sea area with complicated topography. To solve this problem, this paper proposes a method for semantic segmentation of storm surge based on the three-channel graph attention convolution, called Tri-GACNet, using point cloud data to identify and segment different regions such as storm surge tides and coastal slabs. First, construct the storm surge point cloud dataset. Then, three modules of global feature extraction, local similarity calculation and edge feature enhancement are designed and constructed to enhance the model’s ability to extract multi-scale features of storm surge. Finally, the storm surge point cloud features at different scales are fused in three dimensions: global, local and edge to improve the performance of semantic segmentation of storm surge point cloud data. The experimental results show that Tri-GACNet is able to accurately partition the different regions of the storm surge, thus reducing the impact on the storm surge volume calculation due to the large differences in coastal morphology between different shore sections.