近红外光谱分析技术快速检测藕粉品质

目的 基于近红外光谱技术鉴别不同产地的藕粉样品与检测藕粉的掺假问题。方法 采集不同产地的藕粉样品的近红外光谱，在光谱预处理后采用相关系数法提取特征波长，以提取的特征波长变量构建偏最小二乘法-判别分析(partial least squares discriminant analysis，PLS-DA)、线性判别分析(linear discriminant analysis，LDA)以及支持向量机(support vector machine，SVM)模型，实现对不同产地藕粉的鉴别分析。同时，采集掺假地瓜粉、玉米粉、木薯粉的藕粉样品的近红外光谱，在样品类别已知情况下，运用K-Means聚类分析鉴别3种掺假类型的藕粉样品，在掺假类别未知下，运用基于局部密度判别的聚类算法进行判别。结果 以相关系数法提取的特征波长变量构建的PLS-DA、LDA和SVM三种模型对于不同产地藕粉样品的判别准确率均为100%。对于不同掺假类型的藕粉检测，在样品类别已知情况下，K-Means聚类分析能有效识别出掺假藕粉，识别精度为98.33%。在样品类别未知的情况下，基于局部密度判别的聚类算法可以有效识别出2%掺假率的藕粉样品。结论 近红外光谱技术能实现不同产地莲藕粉的快速鉴别，同时为隐蔽的藕粉掺假鉴别提供一种快速、高效、无损检测的分析方法，为藕粉的质量控制提供一定的理论基础。

Rapid detection of lotus root starch quality based on near infrared spectroscopy

Objective Identification of lotus root starch samples from different geographical origins and detection of adulteration of lotus root starch based on near infrared spectroscopy. Methods The spectra of lotus root starch samples from different geographical origins are collected. The characteristic wavelengths are extracted by correlation coefficient method after spectral preprocessing. Based on the extracted characteristic wavelength variables, the models of partial least squares discriminant analysis (PLS-DA), linear discriminant analysis (LDA) and support vector machine (SVM) are constructed. In order to detect the adulteration of lotus root starch, the spectra of lotus root starch samples adulterated with sweet potato flour, corn flour and cassava flour were collected. When the sample category was known, the K-means clustering analysis was used to identify the three adulterated lotus root starch samples. When the adulteration category was unknown, the clustering algorithm based on local density discrimination was utilized. Results PLS-DA, LDA and SVM models based on the characteristic wavelength variables extracted by the correlation coefficient method have a discrimination accuracy of 100% for lotus root starch samples from different geographical origins. For the detection of different adulterated lotus root starch, when the sample category is known, the K-means cluster analysis can effectively identify the adulterated lotus root starch and the recognition accuracy is 98.33%. When the sample category is unknown, the clustering algorithm based on local density discrimination can effectively identify the lotus root starch samples with 2% adulteration rate. Conclusion The results showed that near infrared spectroscopy is a fast, accurate and non-destructive analysis method to rapidly identify the lotus root starch samples from different geographical origins and to detect the adulteration of lotus root starch. It provided a theoretical basis for the quality control of lotus root starch.