Ideal Point Fusion for Reliable Morphometry of Lunar Regolith Particles

Lunar regolith particle morphology is critical for particle classification and representative sample selection. Morphometric measurements also provide key parameters for discrete element modeling and the development of regolith simulants with controlled particle shapes. However, their accuracy depends directly on segmentation quality. This study investigates whether combining measurements from complementary segmentation models can improve accuracy across morphological descriptors. The fusion leaves the source masks unchanged. The objective is to develop and evaluate Morphology-Aware Ideal-Point Fusion (MIPF). MIPF evaluates each model in three dimensions: region completeness, boundary accuracy, and false-positive suppression. Ideal-point analysis assigns higher scores to models closer to the best observed performance and farther from the worst. MIPF uses descriptor-specific weights for these scores to determine each model’s contribution. Area-based indices place more weight on region completeness, while contour-based indices emphasize boundary accuracy. MIPF then retains the two highest-weighted models and combines their measurements without generating a new mask. In a ground-truth comparison based on dataset-level means for 15 morphological indices, MIPF ranks first for six indices and second for the remaining nine, giving an average rank of 1.60. Its median relative error is 11.56%, compared with 12.96% for U-Mamba, 13.73% for nnU-Net, 15.30% for OffSeg, and 31.18% for SegFormer. Although MIPF does not minimize the error for every index, it achieves the most consistent overall performance and reduces the dependence of aggregate morphometric results on any single segmentation model.