摘要
该研究抽取宁夏固原市、海原县、同心县、中卫市、中宁县、永宁县、平罗县等市(县)饲料原料豆粕样品100批,将样品利用浓度梯度法划分定标集和验证集,用NIRS对定标集进行扫描采集光谱,采用目前广泛使用的偏最小二乘法建立定标模型,并分别经过无预处理、均值中心化、标准正态变量转换、一阶导数、标准正态变量转换结合去趋势校正(SNV+D)预处理光谱,获得经SNV+D处理后得到的豆粕水分、粗蛋白、粗纤维、粗灰分预测决定系数Rp^(2)分别为0.983、0.968、0.883、0.960,相对分析误差RPD均大于3的近红外模型,通过预测模型验证集验证,并对模型预测值与实测值进行U检验,结果为差异不显著(P>0.05)。表明该近红外模型具有较佳预测效果。
100 soybean meal samples were selected from Ningxia Guyuan City,Haiyuan County,Tongxin County,Zhongwei City,Zhongning County,Yongning County and Pingluo County.Then,concentration gradient method was applied to divide the calibration set and validation set.NIRS was used to collect spectrums after scanning the calibration set.The commonly used partial least squares(PLS)was used to construct the calibration model.On the basis,The spectrums were preprocessed by using the methods of no pretreatment,mean value zero-centered,standard normal variable(SNV)transformation,first-order derivative and SNV+D(De-trending).Finally,the determination coefficients of water,crude protein,crude fiber and crude ash in soybean meals were obtained after the treatment of SNV+D.That is,Rp^(2)=0.983,0.968,0.883 and 0.960.The relative percent differences(RPD)were larger than 3.U-test was carried out for the model prediction values and the measured values.The results indicated that there were no significant differences(P>0.05).It has been proved that the NIR model has sound prediction effects.
作者
陈海燕
杨俊华
杨奇
谢荣国
刘维华
刘宁
Chen Haiyan;Yang Junhua;Yang Qi;Xie Rongguo;Liu Weihua;Liu Ning(Ningxia Veterinary Drugs and Fodder Inspection Institute,Yinchuan,Ningxia 750011)
出处
《宁夏农林科技》
2020年第11期1-5,共5页
Journal of Ningxia Agriculture and Forestry Science and Technology
基金
宁夏回族自治区重点研发计划重大(重点)项目“畜禽产品质量安全评估与监控技术应用研究”(2018BBF02002)。
