Ore Crushing Position Study Method Based on Stacked Ore Point Cloud Instance Segmentation

To address the technical challenges in industrial applications involving stacked ores-particularly the difficulties in distinguishing individual particles resulting from mutual occlusion and geometric irregularities, and the limitations of conventional approaches in precisely determining optimal crushing positions-this study develops a multimodal feature fusion framework integrating ore segmentation with intelligent crushing position determination. Instance Segmentation Level: a hierarchical segmentation framework is constructed by integrating super-voxel clustering with 3D point cloud surface concavity-convexity analysis. To address edge segmentation optimization, a curvature-constrained region growing algorithm is introduced. Furthermore, an adhesion-aware concavity-convexity evaluation function is established to achieve precise separation of adherent ores, effectively mitigating the over-segmentation issues inherent in traditional Euclidean clustering methods when handling complex stacked scenarios. Positional Decision-Making Level: we propose a crushing point localization method incorporating multi-scale geometric feature fusion. Poisson surface reconstruction is employed to construct a continuous geometric model of the ore surface. This is combined with an enhanced RANSAC plane detection algorithm to identify optimal crushing planes, followed by a comprehensive analysis to determine crushing direction vectors. Experimental results demonstrate that the method can effectively segment individual ores in complex stacking scenarios and optimize crushing position determination based on geometric features, providing reliable technical support for automated crushing operations.