装船作业中基于深度神经网络的激光雷达点云语义分割研究

Deep Neural Network-Based Semantic Segmentation of LiDAR Point Clouds for Ship Loading Operations

摘要: 高效处理与分割激光雷达采集的点云数据是无人装船作业中的关键挑战。针对传统分割方法难以直接对原始雷达点云进行有效分割的问题，本文设计了一种基于深度神经网络的语义分割系统，适用于装船机场景下的点云数据处理。该系统通过“人工小规模标注与半自动大规模标注相结合”的策略构建标注数据集，并依托深度神经网络模型实现点云的语义分割。实验结果表明，该研究适用于大规模、显存受限的装船机场景点云分割任务。

Abstract: Efficient processing and segmentation of point cloud data acquired by LiDAR is a critical challenge in unmanned ship loading operations. To address the difficulties traditional segmentation methods face in directly segmenting raw LiDAR point clouds, this paper designs a deep neural network-based semantic segmentation system tailored for point cloud processing in ship loader scenarios. The system builds a labeled dataset using a combined strategy of small-scale manual annotation and large-scale semi-automatic annotation, and performs semantic segmentation of the point clouds with a deep neural network model. The experimental results show that this method is suitable for large-scale, memory-constrained point cloud segmentation tasks in ship loader environments.