基于主成分分析和人工神经网络的近红外光谱大豆产地识别

为了准确、快速地识别大豆产地，通过近红外光谱技术（NIRS）结合主成分分析（PCA）和人工神经网络技术（ANN）研究不同国家大豆内含特征，建立进口大豆产地识别模型。采用箱型图校正法，剔除阿根廷、巴西、乌拉圭、美国等4个国家166组大豆样本中12组异常样本。采用多元散射校正（MSC）、标准正态变量（SNV）、Savitzky-Golay（SG）平滑滤波等方法进行光谱数据预处理，结果表明，采用SG（3）平滑结合MSC预处理效果最好。主成分分析表明，前10个主成分的累积贡献率达到99.966%。选取主成分分析得到前10个主成分为输入向量，4个产地作为目标向量，分别采用支持向量机（SVM）、邻近算法（KNN）与人工神经网络法（ANN）建立识别模型。结果表明，采用BP-ANN建模效果最好，总体测试集准确率为95.65%，其中阿根廷准确率为100%，巴西准确率为100%，乌拉圭准确率为80%，美国准确率为100%，该模型能够实现对进口大豆生产国别的识别。

Soybean Origin Identification Based by Near-Infrared SpectrumBased on Principal Component Analysis and Artificial NeuralNetwork Model

In order to identify the geographical origin of soybeans accurately and quickly,the characteristics of soybean in different countries were studied by near infrared spectroscopy(NIRS),principal component analysis(PCA)and artificial neural network(ANN),and the identification model of imported soybean producing areas was established.The nearinfrared reflectance spectrums of the 166 soybeans from Argentina,Brazil,Uruguay and the United States had been collected,then the twelve outliers of NIRS were eliminate by using a box-plot graph.The original spectral data was processed by means of multiplicative scatter correction(MSC),standard normal variate(SNV),Savitzky-Golay(SG),etc.It was gotten the optimal result by the preprocessing method based on the smoothing treatment in SG(3 points)with MSC.The PCA was used to compress the NIRS,the analysis results showed that the cumulative variance contribution of PC1 to PC10(the first ten principal components)were 99.966%.The first 10 principal components obtained by principal component analysis were selected as input vectors,and four producing areas were selected as target vectors.The recognition models were established by support vector machine(SVM),neighbor algorithm(KNN)and artificial neural network(ANN).The results showed that the BP-ANN model was the best,and the overall discrimination accuracy for the test set was 95.65%by ANN model,and the discrimination accuracy for soybean samples from Argentina,Brazil,Uruguay and the United States were 100%,100%,80%and 100%,respectively.The ANN model can identify the origin of soybeans imported from different countriesrall discrimination accuracy for the test set was 95.65%by ANN model,and the discrimination accuracy for soybean samples from Argentina,Brazil,Uruguay and the United States were 100%,100%,80%and 100%,respectively.The ANN model can identify the origin of soybeans imported from different countries.