Moisture determination in Cyperus esculentus by using visible/near-infrared transmission spectroscopy coupled with sample distribution optimization

Transmission spectroscopy is being increasingly employed for nondestructive evaluation of agricultural product quality. However, the predictive performance of this technique largely depends on the representativeness of the selected calibration and validation samples. A self-designed transmission spectroscopy device was developed for evaluation of Cyperus esculentus, and a sample selection algorithm was employed to optimize dataset partitioning. The device enabled stable spectral acquisition, while the algorithm effectively reduced sample redundancy and improved model generalization. The experimental results demonstrated that in comparison with conventional methods, the proposed method achieved superior predictive performance for measurement of the moisture content in Cyperus esculentus (coefficient of determination for the calibration set [R_c²] = 0.9701, root mean square error of calibration [RMSEC] = 0.0158; coefficient of determination for the validation set [R_v²] = 0.9554, root mean square error of validation [RMSEV] = 0.0174). This study highlights the potential of combining device innovation with advanced sample selection strategies to improve spectral modeling in agricultural and food research.