一种基于激光雷达强度-几何融合的航天器表面材料识别方法及系统 A LiDAR Intensity–Geometry Fusion Method and System for Spacecraft Surface Material Recognition

This invention discloses a LiDAR intensity–geometry fusion method and system for spacecraft surface material recognition, targeting tasks such as on-orbit inspection, autonomous grasping, obstacle-avoidance planning, and thermal-contact constraint determination. To overcome geometric homogeneity at meter-scale sensing distances (where shape alone cannot separate functionally distinct parts) and the complex optics of spacecraft coatings (strong specular reflection, low albedo, multi-layer backscattering), the method comprises: S1 scanning with a solid-state LiDAR to acquire point-cloud sequences with 3D coordinates and raw echo intensity; S2 region-of-interest (ROI) cropping and merging consecutive frames into windows of W frames; S3 range–incidence-angle correction (I_corr = I_raw·d²/cosθ) and normalization of echo intensity, plus a validity mask that retains empty-echo grid slots at grazing angles; S4 spatial blocking (20 cm blocks, 10 cm stride) with farthest-point-sampling resampling; S5 per-point 5-D features (centroid-centered coordinates, normalized intensity, validity mask) and 6-D global features (log valid-point count, valid-point ratio, intensity mean/std, normalized range and incidence-angle priors); S6 a point-cloud deep network that outputs six material-class probabilities; S7 time-window probability averaging and per-point weighted fusion with argmax to produce a material class map with confidence. Recognized classes include multi-layer insulation (MLI), carbon-fiber-reinforced polymer (CFRP), black anodized aluminum, solar panels, bare aluminum, and white-painted aluminum. Darkroom validation yields ~92.62% patch-level accuracy across all conditions (vs. ~90.45% for an xyz-only baseline), still ~86.77% at 3 m / 60° grazing incidence, with no recording-level misclassification.