To address the challenges in automated dry bulk terminal operations of gantry cranes, where strong spatial coupling between wire ropes and the grab bucket in point clouds leads to blurred boundaries, and on-site installation constraints result in sparse and uneven point clouds, making it difficult for traditional segmentation methods to effectively identify wire ropes, a wire rope recognition method based on LiDAR and an improved PointNet++ is proposed. First, point clouds are acquired using LiDAR, and a dynamic pre-segmentation algorithm based on kinematic constraints is employed to extract the operational area, thereby reducing computational redundancy. Subsequently, a spatial-self-attention (SSA) module is embedded into the PointNet++ framework. The self-attention branch establishes long-range dependencies across points to enhance the axial topological continuity of the wire rope, while the spatial attention branch dynamically weights local geometric features to strengthen the boundary response in the coupled region. Their synergistic effect improves segmentation accuracy in sparse and coupled scenarios. Experiments conducted on a self-constructed gantry crane operation point cloud dataset demonstrate that the improved method achieves a wire rope segmentation IoU of 91.2%, an improvement of 9.4% over the baseline PointNet++. Under adverse weather conditions such as rain, fog, and snow, the IoU consistently exceeds 89.8%. Across various real-world operating conditions, including entering and exiting the hatch and boom slewing, the IoU remains stable between 89.4% and 92.7%, significantly outperforming comparative methods. This effectively solves the problem of precise wire rope segmentation in strongly coupled and sparse point cloud scenarios, providing a reliable perceptual foundation for anti-sway control and closed-loop positioning of grabs in automated terminals.