基于X射线荧光光谱分析技术的进口大豆产地鉴别

Origin Discrimination Method of Imported Soybean Based on X-ray Fluorescence Spectrum

本研究收集了4个国家共166组大豆样本,采用X射线荧光光谱技术测定大豆样本中多元素含量,使用箱型图校正法,剔除4组异常样本。对样本数据集进行主成分分析,结果表明,前5个主成分的累积方差贡献率达到99.67%。前5个主成分为输入向量,4个产地作为目标向量,采用支持向量机(SVM)、RUSBoosted树(RUSBoosted Trees)与人工神经网络法(ANN)建立产地鉴别模型。人工神经网络建模鉴别效果最好,整体鉴别准确率可达95.8%,其中阿根廷、巴西和乌拉圭鉴别准确率均为100%,美国鉴别准确率为85.7%。

In this study, a total of 166 sets of soybean samples from four countries were selected to measure their multi-element content by X-ray fluorescence spectrometry, and then four sets of abnormal samples were excluded using the box plot correction method. Principal component analysis was performed on the sample dataset, which indicated that the cumulative contribution of the first five principal components came to 99.67%. With the first 5 principal components as input vectors and 4 producing places as target vectors, the identification model for place of origin was established using Support Vector Machine(SVM), RUSBoosted Trees and Artificial Neural Network method(ANN). Artificial neural network modeling can get the best identification effect, and the overall identification accuracy rate can reach 95.8%, of which the identification accuracy rate of Argentina, Brazil and Uruguay all is 100%,and that of the United States is 85.7%.