大米蛋白质脂肪总糖水分近红外检测模型研究.docVIP

大米蛋白质、脂肪、总糖、水分近红外检测模型研究 李路1 黄汉英1* 赵思明2 胡月来1 杨素仙1 （华中农业大学工学院1，武汉 430070） （华中农业大学食品科技学院2，武汉 430070） 摘 要 采用近红外光谱技术对大米蛋白质、脂肪、总糖、含水量进行检测。运用经典Kennard-Stone法选取校正集及预测集样本，运用分段小波消噪对光谱进行预处理，通过竞争性自适应重加权采样筛选出与样本性质相关的特征波长，比较偏最小二乘法和BP神经网络法所建立的大米蛋白质、脂肪、总糖、含水量的检测模型。对于大米蛋白质、总糖和水分的检测，两种方法所建立模型的决定系数均大于0.9，相对标准差均小于2.6%，具有良好的精度和稳定性。对于大米脂肪的检测，偏最小二乘模型的性能相对稍好，其决定系数为0.949 5，相对标准差为13.69%关键词 大米 光谱 PLS BP神经网络 CARS 中图分类号：TS201.2 文献标识码：A 文章编号： Research on NIR spectra detection model of Protein, Fat, Total Sugar and Moisture in Rice Li Lu1 Huang Hanying1* Zhao Siming2 Hu Yuelai1 Yang Suxian1 (College of Engineering, Huazhong Agricultural University1, Wuhan 430070) （College of Food Science Technology, Huazhong Agricultural University2, Wuhan 430070)Abstract The Near-Infrared (NIR) spectrum was used to detect the protein, fat, total sugar and moisture content in rice. Kennard-Stone method was used to select the calibration set and prediction set samples. Then the segmented wavelet de-noising algorithm was used to preprocess the spectrum before screening the characteristic wavelength applied the competitive adaptive reweighted sampling (CARS). The detection models of the rice protein, fat, total sugar and moisture content based on the partial least squares (PLS) and the BP neural network were compared to determine an appropriate detecting method. The coefficient of determination (R2) of the models for the rice?protein, total sugar and moisture content detection were 0.9, and the relative standard deviation (RSD) were 2.6%. So all models established by the two methods for the rice?protein, total sugar and moisture content detection were accurate and stable. For detecting the rice fat content, the performance of the PLS model was better whose R2 and RSD value were 0.949 5 and 13.69%. Key Words Rice, NIR spectroscopy, PLS, BP neural network, CARS 蛋白质、脂肪、总糖、含水量是评价大米品质的重要指标[1]。而其传统测定方法存在过程繁琐、条件不易控制、药品消耗大、费时费力等不足。近红外光谱分析技术以其检测过程简单迅速、耗材少、无损、分析重现性好、成本低等优点，在粮食、蔬果、肉制品等方面的检测中得到了广泛的应用[2