近红外光谱技术结合宽度学习系统识别国外奶粉产地

Origin identification of foreign milk powder using near-infrared spectroscopy technology coupled with broad learning system

中文关键词:  奶粉  产地识别  近红外光谱技术  主成分分析  宽度学习系统

英文关键词:milk powder  origin identification  near-infrared spectroscopy  principal component analysis  broad learning system

中文摘要:

      目的 利用傅里叶变换近红外光谱技术结合与宽度学习系统对国外奶粉进行产地识别。方法 采集荷兰、新西兰、澳大利亚、德国、法国、英国和爱尔兰7个国家55个奶粉样品的近红外光谱, 经过数据预处理、主成分分析降低数据维度和特征筛选, 构建基于宽度学习系统(broad learning system, BLS)的奶粉产地快速识别模型。结果 采用多元散射校正加Savitzky-Golay滤波的预处理效果最好, 与未做预处理相比, 准确率提高14.55%, 主成分分析特征数大于38, 识别效果最稳定。对荷兰、新西兰、澳大利亚和欧洲其他产地4类产地识别, 测试准确率达到100.00%, 对样本做7类产地识别, 准确率达到81.81%。相同条件下, 与支持向量机方法对比, 4类产地识别, BLS方法准确率比支持向量机方法高9.10%, 7类产地识别, 两者准确率相同。结论 本研究提出的基于BLS的方法可以较好实现国外奶粉产地识别, 为奶粉产地快速识别提供了新思路。

英文摘要:

      Objective To identify the foreign milk powder’s geographical origin by Fourier transform near-infrared spectroscopy technology coupled with broad learning system. Methods The near infrared spectra of 55 milk powder samples from 7 countries including the Netherlands, New Zealand, Australia, Germany, France, Britain and Ireland were collected. After preprocessing, dimensionality reducing by principal component analysis and feature screening, the data were used as the input of broad learning system (BLS) to establish a fast origin recognition model. Results The preprocessing effects of multivariate scattering correction with Savitzky-Golay filter was the best, and the accuracy was 14.55% higher than that without pretreatment. When the number of principal component analysis features was greater than 38, and the recognition effect was the most stable. For the identification of 4 geographical origins in the Netherlands, New Zealand, Australia and other European origin, the test accuracy was 100.00%, and the recognition accuracy of 7 geographical origins was 81.81%. Under the same conditions, compared with support vector machine, the accuracy of BLS method was 9.10% higher than that of standard normal variate transform method for 4 types of origins recognition, and the accuracy of 7 types of origin recognition was the same. Conclusion The proposed method based on BLS can better realize the origin recognition of the foreign milk powder. It provides a new idea for the fast origin recognition of milk powder.