基于近红外无损快速检测淀粉品质的研究

Study on Rapid and Undamaged Determination of Starch Quality Based on Near-infrared Spectroscopy

实验选择最优的光谱预处理方法和光谱范围,运用近红外光谱分析中的聚类分析法和偏最小二乘法(PLS)分别进行了定性和定量分析研究。结果表明,采用聚类分析法可准确地将样品分为两类,同时采用预测集的样品验证所建模型的准确性,预测准确率达90%以上。采用偏最小二乘法(PLS)所建的定量分析模型的相关性较高,预测相关系数和预测均方根误差均符合要求。

In the study,it chooses the best method for spectrum preprocessing and the optimal spectral range.It also uses clustering methodology and PLS to do quantitative and qualitative analysis,respectively.The result shows that the samples are classified into two groups correctly by using clustering methodology.In the meantime,by using the test set samples,it is safe to test the accuracy of the model.The predict accuracy is over 90 percent of all.The quantitative analysis model which is established by PLS has a high related performance.The prediction correlation coefficients and root-mean-square prediction error are both up to requirements.