摘要
采用近红外光谱分析技术无损鉴别鲜枣品种和测定其可溶性固形物含量。对3个不同品种的鲜枣进行光谱分析,各获取30个样本数据。采用平滑法和多元散射校正方法对样本数据进行预处理,再用主成分分析法对光谱数据进行聚类分析并获得各主成分数据。将样本随机分成75个建模样本和15个预测样本,将建模样本的主成分数据作为BP神经网络的输入变量,鲜枣品种和可溶性固形物含量作为输出变量,建立3层人工神经网络鉴别模型,并用该模型对15个预测样本进行预测。结果表明,在阈值设定为±0.17的情况下该模型对预测集样本品种鉴别准确率达到100%,可溶性固形物含量预测值与实测值相对偏差小于10%。
A non-destructive detection method based on near infrared reflectance （NIR） spectroscopy and chemometrics was put forward for discriminating varieties and detecting soluble solids content （SSC） of fresh jujube. A FieldSpec 3 spectroradiometer was used for collecting 30 sample spectra data of the three kinds of jujube separately. Then principal component analysis was used to process the spectral data after pretreatment. Six principal components （PCs） were selected based on accumulative reliabilities, and these selected PCs would be taken as the inputs of the three-layer back-propagation artificial neural network （BP- ANN）. A total of 90 jujube samples were divided into calibration set and validation set randomly, the calibration set had 75 samples with 25 samples of each variety and the validation set had 15 samples with 5 samples of each variety. The BP -ANN was trained using samples in calibration set. The optimal three-layer BP - ANN model with 6 nodes in input layer, 10 nodes in hidden layer, and 2 nodes in output layer would be obtained. Then this model was used to predict the sample in the validation set. The result show that a 100% recognition ration was achieved with the threshold predictive error ± 0.17, the bias between predictive value and standard value was lower than 10%.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2009年第4期139-142,共4页
Transactions of the Chinese Society for Agricultural Machinery
基金
山西省科技攻关项目(2007031109-2)
关键词
鲜枣
近红外光谱
无损检测
主成分分析
BP神经网络
Fresh jujube, NIR spectroscopy, Non-destructive detection, Principal componentanalysis, BP neural network
