Calibration Coverage and Repeatability in LIBS and Hyperspectral Imaging Fusion for Starch-Adulterated Chuanbei Powder

Quantifying fillers in medicinal plant powders supports quality control when grinding obscures visual identification. We evaluated near-infrared hyperspectral imaging (HSI) and laser-induced breakdown spectroscopy (LIBS) for corn starch quantification in Chuanbei powder. Both modalities measured aliquots from the same blends, and the 20%, 30%, and 40% formulations were withheld in turn. Under matched calibration, mean absolute errors (MAEs) were 3.32 percentage points for HSI, 11.94 for LIBS, and 7.56 for prediction averaging. Expanding calibration reversed the HSI error at 40%, enabling partial cancellation with the LIBS error. Row-specific LIBS calibration reduced the mean within-formulation standard deviation of fused predictions from 6.97 to 3.83 percentage points, while formulation-mean MAE changed from 6.87 to 6.80. Of 28 joint settings screened, one local calibration yielded a lower MAE than the corresponding HSI and LIBS curves (1.79 versus 2.25 and 2.97 percentage points). HSI kernel regression achieved 0.98 on the same local grid. Spectral substitutions showed that the measured LIBS calibration and test spectra contributed to the improvement at 40%. These findings clarify how to evaluate fusion for medicinal powder quality control by distinguishing prediction repeatability from concentration accuracy.