The particle size and non-uniformity of solid particle samples directly affected the acquisition of samples spectral data. In this study, different particle size samples and mixed samples with two different particle size grade samples were applied to investigate the impact of particle size and non-uniformity on near-infrared (NIR) spectroscopy detection of moisture content in coco-peat substrate. Simultaneously, the pretreatment effect and capability of MSC and SNV on spectral scattering were studied. The results showed that the prediction accuracy of the spectral detection model for moisture content in coco-peat substrate was higher with smaller particle size or smaller non-uniformity; MSC and SNV had better effects on improving the accuracy of prediction models with small particle size samples or small non-uniformity samples; the optimal moisture content spectral prediction models for single particle size grade samples and mixed samples of two particle size grades could be respectively established when SNV was adopted to pretreat the samples spectral data of A and A-B. And the corresponding correlation coefficients of the two optimal models were 0.9964 and 0.9959 for calibration set, 0.9959 and 0.9957 for prediction set, respectively; the root mean square errors were 1.2299% and 1.0090% for calibration set, 1.2820% and 1.0723% for prediction set, respectively; the ratios of prediction to deviation were 10.83 and 10.51, respectively. The prediction model with the minimal spectral scattering influence could be built by combining physical pretreatment with spectral pretreatment. This study would provide a reference for reducing the scattering effect of solid particle sample spectral detection.
Read full abstract