摘要
为了利用近红外光谱(NIRS)技术预测重庆地区狼尾草属(Pennisetum)牧草的品质,本研究以重庆地区栽培利用的10个狼尾草属牧草品种的165份样品为试验材料,基于支持向量机(SVM)算法与偏最小二乘(PLS)算法分别建立狼尾草属牧草水分(M)、粗蛋白质(CP)、粗纤维(CF)、酸性洗涤木质素(ADL)、酸性洗涤纤维(ADF)、中性洗涤纤维(NDF)、粗灰分(Ash)含量的NIRS预测模型。结果显示:SVM算法提高了预测模型的预测精度。在对7项营养品质指标的预测中,SVM模型获得预测效果良好[相对预测误差(RPD)≥3.0]的有5个指标,预测效果可以接受(2.5<RPD<3.0)的有2个指标,没有预测效果不可接受(RPD≤2.5)的指标;PLS模型获得预测效果良好(RPD≥3.0)的仅有2个指标,预测效果可以接受(2.5<RPD<3)的只有1个指标,预测效果不可接受(RPD≤2.5)的有4个指标,表明该模型的预测效果不可接受,不能应用于NIRS定量分析。基于SVM算法建立的狼尾草属牧草品质预测模型总体上优于基于PLS算法建立的预测模型,其部分指标的预测决定系数大于0.80。综上可知,与PLS算法相比,SVM算法对于NIRS这种非线性数据具有良好的拟合能力,可应用于狼尾草属牧草品质分级。
To predict the quality of forages from Pennisetum in Chongqing by near-infrared spectrum(NIRS)technique,a total of 165 samples of 10 cultivars of Pennisetum cultivated in Chongqing as research materials,while 70%samples were selected as calibration set and 30%samples as verification set,to establish the NIRS prediction models of moisture(M),crude protein(CP),crude fibre(CF),acid detergent lignin(ADL),acid detergent fibre(ADF),neutral detergent fibre(NDF)and crude ash(Ash)contents of forages from Pennisetum based support vector machine(SVM)algorithm and partial least square(PLS)algorithm,respectively.The results showed that the SVM algorithm improved the prediction accuracy of prediction models.Among the seven nutritional quality indexes,SVM model obtained five indexes with good prediction effect[relative prediction error(RPD)≥3.0],two indexes with acceptable prediction effect(2.5<RPD<3.0),and there were no indexes with unacceptable prediction effect(RPD≤2.5),while PLS model obtained only two indexes with good prediction effect(RPD≥3.0),there is only one index with acceptable prediction effect(2.5<RPD<3.0),and there were four indexes with unacceptable prediction effect(RPD≤2.5),indicating that the prediction effect of the model is unacceptable and could not be applied to NIRS quantitative analysis.Overall,the quality prediction model of forages from Pennisetum established by SVM algorithm was better than the prediction model established by PLS algorithm,and the prediction determination coefficients of some indexes were greater than 0.80.To sum up,SVM algorithm has good fitting ability for the nonlinear data of NIRS,which can be applied to forage grass quality classification of Pennisetum.
作者
朱瑞芬
徐远东
刘畅
孙万斌
姚博
游斌杰
陈积山
ZHU Ruifen;XU Yuandong;LIU Chang;SUN Wanbin;YAO Bo;YOU Binjie;CHEN Jishan(Institute of Pratacultural Science,Chongqing Academy of Animal Sciences,Chongqing 402460,China;Pratacultural Engineering and Technology Research Center of Chongqing,Chongqing 402460,China)
出处
《动物营养学报》
CAS
CSCD
北大核心
2024年第6期3984-3994,共11页
CHINESE JOURNAL OF ANIMAL NUTRITION
基金
国家重点研发计划项目(2022YFD13008005)
重庆市财政专项资金项目(22538C)
饲草生产力提升与品质智能评价系统开发(23201F)。
关键词
近红外光谱
狼尾草属牧草
常规营养成分
支持向量机
偏最小二乘
near infrared spectrum
forages from Pennisetum
conventional nutrients
support vector machine
partial least square
