【作者】 陆月青；

【导师】 梁明振； 谢梅冬；

【作者基本信息】 广西大学 ， 动物营养与饲料科学， 2007， 硕士

【摘要】 饲料原料的品质是保证饲料产品质量与安全的根本，为了能快速有效地控制原料的品质，本课题采用近红外光谱(NIRS)快速检测的方法，收集了49个纯鱼粉样品，按1％、5％、9％的比例分别向纯鱼粉样品中掺入了尿素、豆粕、麦麸和菜粕，制成12个假鱼粉，市面上售有的4个掺假鱼粉，50个豆粕样本和50个麦麸，分别在950～1650nm范围内进行近红外光谱扫描，采用不同的光谱处理方法分别进行了鱼粉的掺假判别、豆粕尿素酶活性及鱼粉、豆粕、麦麸常规化学成分的近红外光谱定量分析模型的建立，结果如下：1、先用主成分回归(PCR)对56个鱼粉样品(44个纯鱼粉及12个自制的掺假鱼粉)进行分析，可以明显地区分纯鱼粉与掺假鱼粉。由此可知，近红外光谱法可作为鉴别鱼粉的一项新技术。再在此基础上，采用PCR结合马氏距离(Mahalanobis)法，用44个纯鱼粉样品创建了近红外光谱模型，验证结果表明，采用标准化(SNV)处理方式建立的模型在9主成分数下对验证集样品进行预测，误辨数为0，验证效果最好。2、采用一阶导数(SG1)+附加散射(MSC)+中心化(mean center)预处理方法，用偏最小二乘法(PLS)建成了豆粕尿素酶活性测定的校正模型，其校正决定系数R~2为0.916，定标标准差(SEC)为0.045，外部验证决定系数R~2为0.926，预测标准差SEP=0.035，测量重复变异系数为0.09。结果基本达到了定量分析的要求。3、用偏最小二乘(PLS)定标方法，结合中心化、多元散射、导数处理的方法，对豆粕样品化学成分建立了的定标模型，定标集化学分析值与NIRS预测值之间的决定系数R~2和标准差RMSEC分别为：0.9695和0.117(水分)，0.9755和0.338(粗蛋白)，0.9507和0.271(粗脂肪)，0.845和0.076(粗灰分)。0.9384和0.267(粗纤维)。用验证集样本对NIRS定标模型进行了检验，其预测值与化学分析值之间的决定系数R~2和标准差RMSEP分别为：0.9327和0.115(水分)、0.9533和0.313(粗蛋白)、0.9847和0.221(粗脂肪)、0.8541和0.078(粗灰分)、0.9009和0.248(粗纤维)。除粗灰分外，其它成分均达到了定量分析的要求。用同样的方法对鱼粉各化学成分建立了定标模型，定标集化学分析值与NIRS定标模型预测值之间的决定系致R~2和交互验证标准差RMSECV分别为：0.9559和0.216(水分)，0.9651和0.386(粗蛋白)，0.9421和0.288(粗脂肪)，0.8899和0.249(钙)，0.9553和0.085(磷)，0.9235和0.135(盐分)。用验证集样品对NIRS定标模型进行了检验，其预测值与化学分析值之间的决定系数R~2和标准差RMSEP分别为：0.9395和0.314(水分)，0.908和0.827(粗蛋白)，0.9101和0.613(粗脂肪)，0.8353和0.474(钙)，0.83和0.294(总磷)，0.9238和0.393(盐分)。除钙、磷验证效果稍差之外，其它成分均达到了定量分析的要求。用相同的处理方法对麦麸样品各化学成分建立了定标模型，定标集化学分析值与NIRS定标模型预测值之间的决定系数R~2和交互验证标准差RMSECV分别为：0.9701和0.101(水分)，0.9572和0.123(粗蛋白)，0.9304和0.109(粗脂肪)，0.9814和0.129(粗纤维)，0.9582和0.105(粗灰分)。用验证集样品对NIRS定标模型进行了检验，其预测值与化学分析值之间的决定系数R~2和标准差RMSEP分别为：0.9477和0.142(水分)，0.9517和0.148(粗蛋白)，0.8935和0.133(粗脂肪)，0.9579和0.161(粗纤维)，0.8833和0.114(粗灰分)。各成分定标结果均达到了定量分析的要求。更多还原

【Abstract】 Materials of feed are the base of the feed, and their qualities ensure the qualityand safety of feed products. In order to control the quality of the materials rapidlyand effectly, the method of near infrared reflectance spectroscopy （NIRS） technologywere appplied in this study. 49 true fish meal samples, 4 adulterated fish meal forsale, 50 soybean meal samples and 50 wheat bran samples were collected and delib-erately adulterated with urea, soybean meal,wheat bran and rapeseed meal respectivelyin 1%, 3%, 5% proportion to make 12 adulterated fish meal samples. The sampleswere scaned at the NIRS region 950-1650 nm respectively.The calibration models todetect whether the fish meal samples were adulterated and to predict the activity ofsoybean meal and conventional chemical compositions in fish meal ,soybean meal andwheat bran were developed using different methods to handle the NIR spectrograms ofthe samples. The results were as follows:1. Using the method of PCR to handle the NIR spectrograms of 56 fish mealsamples,we can discriminate true fish meal samples from adulterated fish meal sampleseasily.Therefore, the NIRS technique can be used as a new technique to detect quicklywhether the fish meal samples adulterated or not. Whereafter, the calibration modelwhich was developed with 44 true fish meal samples was established by the meanPCR combine with Mahalanobis distance. The result of validation showed that themodel which developed under the nine pinecipal factors by SNV was the best one,the number of false discrimination was zero.2. The universal calibration model for rapid estimation of the activity of soybean-meal was established by using the partial least square regression（PLS） and fist derivat-ive+multivariate scattering correction+mean center.The R²_cal and SECV was 0.916and 0.045.The R²_val and SEP was 0.926 and 0.035.The coefficient of variation（CV） ofrepeatability was 0.09.3. The universal calibration models for rapid estimation of conventional chemicalcompositions in soybean meal samples were established by using the partial least squ-are regression （PLS） combine with the methods of mean center, multivariate scatteringcorrection and derivative.The coefficient of determination in calibration （R²） and RM-SEC of moisture, crude protein, crude fat,crude ash,crude fiber in soybean meal were 0.9695, 0.9755, 0.9507,0.845, 0.9384 and 0.117, 0.338, 0.271, 0.076, 0.267 respectively.And the coefficient of determination in validation and RMSEP of moisture, crudeprotein, crude fat, crude ash,crude fiber in soybean meal were 0.9327, 0.9533, 0.9847,0.8541, 0.9009 and 0.115, 0.313, 0.221, 0.078, 0.248 respectively.Apart from the crudeash, the other calibrations could meet the required for quantitative analysis.The universal calibration models for rapid estimation of conventional chemicalcompositions in fish meal samples were established by using the same mean. Thecoefficient of determination in calibration （R²） and RMSEC of moisture, crude protein,crude fat,calcium,total phosphor and salt in fish meal samples were 0.9559, 0.9651,0.9421,0.845, 0.8899, 0.9553, 0.9235 and 0.216, 0.386, 0.288, 0.249, 0.085, 0.135Respectively. And the coefficient of determination in validation and RMSEP of mois-ture, crude protein, crude fat,calcium,total phosphor and salt in fish meal sampleswere 0.9395, 0.908, 0.9101, 0.8353, 0.83, 0.9238 and0.314, 0.827, 0.613, 0.474,0.294, 0.393 respectively. Apart from the calcium and total phosphor, the other calib-rations could meet the required for quantitative analysis.The universal calibration models for rapid estimation of conventional chemicalcompositions in wheat bran samples were established by using the same mean. Thecoefficient of determination in calibration （R²） and RMSEC of moisture,crude protein,crude fat, crude fiber, crude ash in wheat bran samples were 0.9701, 0.9572, 0.9304,0.9814, 0.9582 and 0.101, 0.123, 0.109, 0.129, 0.105 respectively. And the coefficientof determination in validation and RMSEP of moisture, crude protein, crude fat, crudefibre, crude ash in wheat bran samples were 0.9477, 0.9517, 0.8935, 0.9579, 0.8833and 0.142, 0.148, 0.133, 0.161, 0.114 respectively.All of the calibrations could meetthe required for quantitative analysis.更多还原