Offshore wind power underwater point cloud data segmentation and target detection method and system

The invention relates to an offshore wind power underwater point cloud data segmentation and target detection method and system. The method comprises the following steps: firstly, carrying out data preprocessing, including voxelization, uniform sampling, multi-scale layered division, random sampling and local feature aggregation, so as to extract a low-dimensional feature vector; in the network structure, the points of each sub-region undergo a multi-layer self-attention mechanism and non-local module fusion feature vectors; and outputting the semantic tag and the instance offset by using the rendering branch at the same time. And in the loss function part, semantic loss and offset loss are combined for optimization. During training, an AdamW optimizer, a learning rate scheduler and a data enhancement technology are adopted, and the robustness and generalization ability of the model are improved. According to the method, semantic and instance segmentation of the point cloud can be effectively carried out, and