Abstract 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.

中文翻译：

---

椰子泥炭基质含水量的定量 NIR 光谱测定：粒度和不均匀性的影响

摘要 固体颗粒样品的粒径和不均匀性直接影响样品光谱数据的获取。在本研究中，使用不同粒径的样品和两种不同粒径等级样品的混合样品，研究粒径和不均匀性对近红外 (NIR) 光谱检测椰子泥炭基质中的水分含量的影响。同时研究了MSC和SNV对光谱散射的预处理效果和能力。结果表明，粒径越小或不均匀性越小，该光谱检测模型对椰糠基质中水分含量的预测精度越高；MSC和SNV对提高小粒径样本或小非均匀样本预测模型的准确性有较好的效果；采用SNV对A、AB样品光谱数据进行预处理，可以分别建立单一粒径级样品和两种粒径级混合样品的最优含水率光谱预测模型。并且两个最优模型对应的相关系数对于校准集分别为0.9964和0.9959，对于预测集分别为0.9959和0.9957；校准集的均方根误差分别为 1.2299% 和 1.0090%，预测集的均方根误差分别为 1.2820% 和 1.0723%；预测与偏差的比率分别为 10.83 和 10.51。物理预处理与光谱预处理相结合，可以建立光谱散射影响最小的预测模型。